Track new features, improvements, and bug fixes across Google Antigravity
surfaces, including Antigravity 2.0, Antigravity CLI, the Antigravity SDK, and
Antigravity IDE.
New versions roll out gradually and may take a few days to reach all users.
This release introduces a Customizations tab and marketplace for discovering and installing plugins and includes overall improvements to the sidebar and chat experience.
Improvements(14)
You can now discover, install and manage plugins within the Customizations tab, with curated shelves for development tools and workspace integrations.
You can now switch the sidebar to show only archived conversations from the Display Options menu and filter by archived conversations on the Conversation History page, with archive folder icons marking archived projects.
Terminal commands in permission requests are now syntax highlighted, making it easier to read what the agent wants to run before approving it.
Terminal command steps in the conversation now stay collapsed by default while running and after completing, expanding automatically only when waiting for your approval.
Slash commands provided by plugins now show their plugin name in the / menu, can be filtered by typing the plugin name, and display a hover card with the plugin name and full description.
Images generated by the agent now stay visible at the bottom of the turn alongside the response text instead of collapsing into the completed steps section.
The token usage breakdown now shows rules and other customizations as separate bars against their own budgets, and flags any rules or customizations that were left out or included by path only after going over budget.
You can now delete any project directly from the overflow menu in its sidebar header.
Press Option+Command+P on macOS or Alt+Shift+P on Windows and Linux to pin or unpin the active conversation.
The startup loading screen now shows a pulsing Antigravity logo in place of the loading dots and text.
Arrow-key navigation in the command palette now wraps around between the first and last rows.
When the agent asks you a question and an answer choice is truncated, hovering over it now shows the full text.
Starting a conversation in a new workspace renders the workspace in the sidebar immediately.
Grouping conversations by workspace in the sidebar now groups by folder path, labels each group with its folder name, and hides empty headers.
Fixes(15)
Fixed an issue where Google Sign-In could fail with a browser security error when signing in from the app or the built-in browser.
Fixed an issue where scrolling up in long conversations could stall after collapsed tool steps, unloading earlier turns could jump the scroll position, and page bounds could collapse to zero.
Fixed an issue where copying text from dark mode and pasting it into rich-text editors such as Google Docs or Gmail carried over dark background colors and light text.
Fixed an issue where closing the app during enterprise sign-in before choosing a license or project left an incomplete sign-in saved the next time you opened it.
Fixed an issue where cancelling a new conversation while it was still being created showed a Conversation unavailable notification a few seconds later.
Fixed an issue where pressing Ctrl+Left or Ctrl+Right on Windows and Linux jumped across the entire prompt when the message box started with an @ mention or slash command chip.
Fixed an issue where pressing Up or Down inside a multiline prompt navigated prompt history instead of moving the cursor between lines unless already at the start or end.
Fixed an issue where pasting a message that contained a slash command sent only the command name to the agent instead of its full instructions.
Fixed an issue where keyboard shortcuts such as Ctrl+/ and Ctrl+M were not routed to the focused pane in split-pane layouts.
Fixed an issue where choosing a sort order in Display Options did not apply if conversations had previously been reordered by dragging.
Fixed an issue where open files were duplicated in @ mention search results and folder paths with special characters were displayed with percent encoding.
Fixed an issue where files in .git, .env, and .vscode directories could be modified without a permission prompt under Default and Request Review permission presets.
Fixed an issue where Git worktrees created by subagents were saved in conversation directories and misclassified as artifacts.
Fixed an issue where background file viewer updates and menus could steal focus while the application window was in the background.
Fixed an issue where copying text containing entity pills dropped the pill labels from the copied text.
Type /plan and the agent drafts a plan you can read and edit before it writes any code. The Agent setting that governs this is now called Plan Review Policy, and you can set it to review every plan, review only when the agent judges it worthwhile, or skip review entirely.
Improvements(11)
Reveal in Finder for projects. Projects in the sidebar have a new Reveal in Finder entry (Show in File Explorer on Windows) in their overflow menu, opening the project folder in your system file manager. Multi-folder projects open one window per folder.
Toggle theme from the command palette. A new "Toggle Between Light/Dark Themes" command flips the theme without a trip to Settings. On System, the first toggle pins the opposite of whatever your OS resolved to.
Custom agents can declare their own hooks. A Markdown agent can list hook files in its front matter with a hooks: entry. Paths may be relative to the agent, relative to the project, or absolute.
Save Video. Right-click a video anywhere it appears — in the chat, the file viewer, a browser recording, or the full-screen preview — and choose Save Video.
Rules no longer crowd out everything else. Rules get their own 20,000-token budget, so a large rules file stops pushing your skills, subagents and MCP tools out of the agent's context. A rule over budget is listed by file path and description for the agent to open when relevant, instead of being cut short.
Per-project configuration moved. Repository customization settings now load from /.gemini/config.json. The older .agents/settings.json is no longer read — move any personal_customization_dir entry into the new file.
Paste properly into the chat box. Right-clicking the chat input and choosing Paste handles images and rich text, keeping links as markdown, instead of only plain text. Pasting over a selected attachment no longer drops the pasted text.
Slash commands and plugins read like products. The pill inserted when you pick a slash command shows the skill's proper name and logo instead of its raw command name, and the built-in Google plugins show readable names — Chrome DevTools, Google Maps Platform, Gemini API, Android CLI, Data Agent Kit, Modern Web Guidance, Google Antigravity SDK.
Fewer link cards you did not ask for. The agent surfaces a link card in the conversation only when there is something to open or act on, such as a review it created, a document it edited, or a local server it started. Links it merely consulted are bookmarked to the sidebar quietly.
Sidebar polish. Hovering a conversation highlights the whole row rather than just the text, and a sidebar that closed itself because you narrowed the window reopens when you widen it again. A sidebar you closed yourself stays closed.
.cppm files are syntax-highlighted as C++ and get the code file icon.
Fixes(10)
Rewinding a conversation no longer restores out-of-date files. The snapshot search used to sweep backward from the point you rewound to, so it could hand back a state from earlier in the conversation. It now takes the first snapshot at or after that point and never an earlier one, and replays the steps when no such snapshot exists.
Large files and diffs no longer freeze the app. Syntax highlighting and line diffing run off the main thread, and files over 100 KB fall back to plain text.
Custom rules, workflows and agents load on large projects. They no longer report a load failure when scanning your project folders takes more than three seconds; the timeout is now fifteen seconds, matching what skills already had.
/btw keeps your agent. A side question runs with the same agent and settings as the conversation you asked it in, instead of falling back to the default agent.
Saving media works for more formats. Saving an image or video from the media menu now works for SVG files and for videos whose type carries codec details; those saves used to fail silently. Files saved from inline data get the right extension.
Maths renders in more cases. Display maths written with the $$ fence attached to the formula on the same line now renders, and no longer swallows the markdown that follows it. Row spacing like \\[12pt] inside a cases block no longer turns into a stray dollar sign, inline maths followed immediately by a digit stays maths, and comma-separated numeric spans no longer swallow the prose after them.
Cut and Copy in the message box are available only when the selected text is inside the message box, and act on the message box itself.
Right-clicking an image caption or an enlarged image no longer pops open the thumbnail's own menu. Images, video and audio whose address carries a query string display inline instead of falling back to a plain link.
The service announcement banner no longer sits on top of the window controls or pushes the breadcrumbs down, and the main title bar stays visible whenever a single conversation pane is open.
Renaming a conversation no longer has the pin, archive and overflow buttons covering the end of a long title.
Windows Subsystem for Linux, Office attachments, and live subagent cards
Connect to a Windows Subsystem for Linux distribution from Application Settings, and drop Word, Excel, and PowerPoint files straight into a prompt. Subagents now appear as live cards you can follow and stop while they work. This release also adds right-click copy and save for images.
Improvements(12)
Added a Windows Subsystem for Linux (WSL) section to Application Settings on Windows, so you can connect to an installed WSL distribution or switch back to your local Windows environment. When connected, the title bar shows the active distribution name.
You can now attach Microsoft Office documents (Word, Excel, and PowerPoint) to a prompt by dragging and dropping them into the message box, and Gemini models read them natively instead of falling back to opening the stored file.
Subagent invocations now appear as dedicated cards in the conversation stream with live running, waiting, and completed status, a hover control to stop a running subagent, and one-click navigation to its conversation in the side pane.
Prompt actions in interactive widgets and HTML artifacts now fill in and focus the message box instead of sending automatically, so you can review or edit the prompt before submitting it.
Right-click an image in an attachment, a preview modal, or an earlier message to Copy Image or Save Image.
SVG images can now be copied to the clipboard and pasted into other applications.
The loading indicator now shows what the agent is doing while it prepares a response, including tool initialization progress and a live countdown with attempt count when a model request is retried.
The timestamp in the agent's message footer now shows when the response finished rather than when it started, matching the duration shown on long-running turns.
Application keyboard shortcuts now take priority over web previews, so shortcuts keep working while a preview has focus.
The Confirm Undo dialog now keeps the file list at a fixed size and scrolls it in place, so Confirm and Cancel stay visible when reverting many files.
Inline pills and icons inside markdown headings now scale with the heading's size and weight instead of rendering small.
Discovering skills and other customizations is now dramatically faster on large projects and network-mounted folders, cutting a startup scan that previously took over a minute to under a second.
Fixes(15)
Fixed an issue where browser sign-in could fail permanently if the app disconnected, reloaded, or the computer went to sleep while you were completing sign-in in your browser.
Fixed an issue where expired or rejected enterprise sign-in credentials stayed cached, leaving the app in a state you could not sign back in from.
Fixed an issue where web search returned no summary when the search summarizer used a reasoning model.
Fixed an issue where reopening an older conversation could re-apply that conversation's earlier file edits to your project and overwrite newer work.
Fixed an issue where a custom icon set on a skill did not appear in the slash command menu, context pills, or step renderers.
Fixed an issue where wide markdown tables could bleed outside a pinned message or show through behind it while scrolling.
Fixed an issue where enlarging an image attached to a restored draft showed a broken image.
Fixed an issue where the Narrow conversation width setting silently fell back to the default width.
Fixed an issue where saving images and other media from a conversation could fail.
Fixed an issue where a "No Models Available" error could flash briefly right after sign-in while models were still loading.
Fixed an issue where screen readers did not announce the name, selected value, or options of dropdown menus in Settings.
Fixed an issue where icons and logos across menus, toolbars, and the sidebar could render blank or lose their color gradients.
Fixed an issue where the conversation history truncation notice appeared at the bottom of the chat instead of at the point where earlier messages were cleared.
Fixed an issue where typing @file: did not show project file suggestions until you typed another character.
Fixed an issue where choosing Paste from the message box's right-click menu did not paste.
Custom agents can now switch off the default prompts and tools, and the agent you pick sticks with you after a reload. This release also adds Page Up, Page Down, Home, End, and Escape for moving around a conversation, speeds up switching conversations in the sidebar, and fixes a batch of settings and conversation state issues.
Improvements(7)
Custom agents can now switch off the default prompt sections and default tools and add back only the tools they need.
Page Up, Page Down, Home, and End now scroll through a conversation, and Escape steps out of the prompt box.
The custom agent you pick in the message box is remembered after a reload, and resets to the main agent if that agent is deleted or turned off.
Dragging a file over the chat now highlights the message box itself, so it is clearer where the file will land.
The project picker above the message box now reads "No Project" when you are working outside of a project, instead of "New Conversation".
Improved performance when switching between conversations in the sidebar.
Bold text in chat now renders at a slightly lighter weight so it sits better with the text around it.
Fixes(16)
Fixed an issue where your saved settings and project list could be overwritten when the app failed to read its state file.
Fixed an issue where permission settings accumulated duplicate entries, which made conversation files grow very large and slowed the app down.
Fixed an issue where starting a new conversation could show a false crash recovery notice.
Fixed an issue where a conversation started without a folder pulled in files and settings from unrelated projects you had open.
A command that is killed when the app restarts is now reported as canceled instead of appearing to finish successfully.
The agent now reports a clear error when it tries to read a binary file it cannot display, instead of returning unreadable content.
A malformed tool call from the model now appears as a short "Invalid tool call" note rather than a failure.
Fixed an issue where an error that arrived without any tool activity in the turn was dropped instead of shown in the conversation.
Fixed an issue where a system message could appear twice, or out of order, in the chat transcript.
Fixed an issue where a turn ending in an error or a scheduled task showed an empty expandable section.
Chat messages now fade out smoothly at the bottom of the conversation instead of being cut off sharply.
The close-window shortcut now closes the dialog on top, such as settings, instead of the conversation behind it, and Escape inside an open menu now closes just the menu rather than the whole dialog.
Fixed an issue where the projects list in settings could not be collapsed again after you expanded it.
Fixed an issue where the side pane's tabs ran underneath the controls in the top-left corner of the window when the pane was maximized and the sidebar was collapsed.
Fixed an issue where selected rows in the file tree and file path menus rendered with a heavy blue fill.
Fixed an issue where sending feedback with large diagnostic logs attached failed, and removed an empty gap in the feedback form.
New Permissions System & Terminal and Git version control for Enterprise
Enterprise and Business accounts can now use the integrated terminal and Git version control in the sidebar. This release also cuts memory use in long conversations and fixes a range of stability and accessibility issues.
Improvements(10)
Enterprise and Business accounts can now use the integrated terminal and Git version control in the sidebar.
Conversations with very long histories now load faster and use less memory.
Removed a redundant scan of your skills, rules, and other customization files that ran on every message, so messages are acknowledged faster.
Settings, artifacts, and terminals opened in their own tab now load faster and no longer flash an empty container first.
The agent's temporary scratch files are no longer watched, synced, or shown in the Artifacts panel and Review pane, so they no longer add noise to either.
Reorganized the conversation menu so Archive and Delete sit together at the bottom behind their own divider, and added Share and Move to Group to the right-click menu.
The command palette now opens over a dimmed backdrop with enter and exit animations, matching the other dialogs in the app.
Renamed the global permissions section on the General settings screen to "Global Permissions" and the inherit option in project settings to "Inherit Global", so the names match the screen they point at.
Pressing the new conversation shortcut on the home screen now toggles between your last-selected project and a standalone conversation instead of staying in standalone mode.
Updated the Vesper dark theme with lilac and pink accents for keywords and operators, and dimmer comments.
Fixes(18)
Fixed an issue where opening Settings, an artifact, or a terminal in its own tab could show a blank screen.
Fixed a crash that could occur when the app failed to read your rule files.
Fixed a memory leak where finished terminal commands held on to their full output for the rest of the session.
Fixed an issue where the app could freeze for several seconds after reconnecting following a period of inactivity, which was most noticeable with a large number of conversations.
Fixed an issue where subagents that had already finished could keep showing as running, which also left follow-up messages queued until restart.
Fixed an issue where a single transient service error could end a session. Transient errors are now retried with backoff for about twelve minutes.
Fixed two issues in the @-mention menu: categories now narrow as you type instead of waiting for search results to return, and pressing Tab no longer overwrites text you already typed.
Fixed an issue where a markdown link pointing at a heading in the same document opened the file viewer at the project root instead of scrolling to that heading.
Fixed an issue where an escaped dollar sign inside a block math expression prevented the formula from rendering.
Fixed an issue where URL artifacts pointing at a Google Drive folder offered "Open Preview" as their primary action, even though Drive folders cannot be displayed in the preview pane.
Fixed an issue where the quota reset time in the model selector was cut off mid-sentence. It now shows a compact duration, with the full text in the tooltip.
Fixed the minimized side question button in the composer toolbar so it shows how many side questions are minimized instead of reading like a mode selector.
Fixed duplicate titles and cramped spacing on the license selection screen.
Fixed an issue where signing in with a personal account after using a business account failed with a project access error.
Fixed an issue where git hooks or global git configuration on your machine could stop the app from saving its internal checkpoints.
Fixed screen reader support for queued messages: queueing a message is now announced along with the queue length and the available actions, and the Send now, Edit, and Delete buttons are labelled.
Fixed screen reader announcements for side questions, which could be repeated or dropped when browsing question history or when an answer arrived in the background.
Fixed an issue where the agent produced Mermaid diagram types the app cannot render, such as Gantt charts and timelines.
Introducing new Documents section and several UX enhancements for organization and control
External files such as Google Drive links, PDFs, and Office documents now have their own Documents section in the sidebar. This release also adds a virtualized viewer for large code and data artifacts, an option to hide whitespace-only changes in diffs, and clearer warnings when a configuration file fails to parse.
Improvements(16)
External files such as Google Drive links, PDFs, and Office documents now appear in their own Documents section in the sidebar above Artifacts.
Added a "Hide Whitespace Changes" option to the Review Changes overflow menu and file diff viewers to filter out whitespace-only edits.
Code and data artifacts like SQL and JSONL files now open in a virtualized viewer with syntax highlighting and line numbers, keeping the app responsive when viewing large files and supporting inline comments.
Commands executed in the sandbox now display a shield badge in the command header that links to sandbox documentation.
Settings displays clear warnings with file paths and a copy button when configuration or project files fail to parse.
Closing a side question now minimizes it to a button in the composer toolbar instead of clearing it. Clicking the button restores the panel so you can review previous side questions without retyping /btw, and a new delete button lets you remove questions you no longer need.
Added a Cancel button and keyboard shortcuts (Ctrl+C on macOS, Ctrl+D on Linux and Windows) to interactive question prompts, letting you stop the agent and return to the chat input without submitting blank answers.
Quote or add selected text to chat using Cmd+L or Cmd+I on macOS, or Ctrl+L and Ctrl+I on Windows and Linux. The text selection popup also adjusts button widths to fit shortcut labels.
Agent scratch files now appear in a dedicated, collapsible Scratch Files section in the side panel rather than mixing into the primary Artifacts list. Turn cards also hide scratch files unless explicitly marked as user-facing.
Unified sidebar section headers and spacing so row heights, collapse controls, and section gaps are consistent between pinned items and projects.
Pin, unpin, and rename conversations directly from the command palette when a conversation is open.
Added an option to archive the active conversation directly from the command palette, and updated archiving to navigate back to the home screen.
Attached review comments now clear immediately from the input box and diff panes when you send a message, and restore if the message fails to send.
Toast notifications stay on screen for 8 seconds instead of 5 seconds by default, giving you more time to read multi-line messages and error details.
Adjusted spacing in the model quota panel and increased its maximum height to eliminate unnecessary scrollbars when viewing quota across multiple models.
Lowered the volume of the task completion sound to match the action-required notification chime.
Fixes(16)
Fixed an issue that could cause the conversation view to crash when rendering messages with repetitive unclosed HTML tags.
Fixed an issue where the model selector could crash if available models failed to load or returned empty groups.
Fixed an issue where screen readers dropped rapid status updates or missed repeated notifications like clipboard confirmations. Very long announcements are now shortened at a sentence boundary with a spoken notice.
Fixed an issue where denying a tool permission request (for example an MCP tool call) removed the step from the conversation. Denied steps now stay visible with a Rejected label.
Fixed an issue where the agent still prompted for permission to read or write artifact files from other projects when file access outside the current project was already allowed.
Removed the usage panel from the model selector for Enterprise and Business accounts, which do not have per-model quotas.
Fixed cramped vertical spacing on cards above the chat input, such as queued messages and input notifications.
Turned off browser and operating system autocorrect, spellcheck, and capitalization in folder path and quick picker inputs so path segments are not changed while typing.
Fixed an issue where hover cards stayed open after switching windows, moving the cursor outside the window, or dragging items.
Fixed an issue where chat autoscroll failed to resume when scrolling near the bottom of a response.
Fixed an issue where quoting chat text containing file links or mentions inserted unwanted line breaks into the quoted text.
Fixed an issue where the scroll-to-bottom button remained visible when viewing trailing whitespace past the latest message, and resolved tooltip flickering during streaming responses.
Custom sidebar widths in the auxiliary pane are now remembered instead of resetting when switching conversations.
Fixed an issue where clicking the copy button on a failed chat turn did nothing instead of copying the error details to the clipboard.
Fixed an issue where .ico files displayed as raw binary text in the diff viewer instead of rendering as visual image previews.
Fixed layout alignment and spacing in the auxiliary pane so the top-right controls line up with the file header icons below.
You can now highlight part of a response and quote it as context for a follow-up prompt, and paid users get the new /boost command for a deeper multi-agent reasoning pass. This release also brings UX improvements across settings, the chat panel, and sidebar navigation.
Improvements(7)
You can now highlight portions of Antigravity responses to quote as context for follow-up prompts.
Introduced the /boost slash command to enhance thinking effort by using a multi-agent reasoning pipeline.
General Settings now shows which of your projects override a setting, with a link to jump straight to that project's settings, and opening Settings opens the active project's settings when overrides are present.
Terminal split layouts are now preserved across window reloads.
The side question panel now has back and forward controls for stepping through the questions you have asked in a session.
You can now send a message with the send button while dictation is still running, instead of having to stop recording first.
Dragging to select a region of an image for a comment now updates the selection immediately upon release.
Fixes(9)
Added drag-and-drop and improved support for attaching audio files to conversations, and fixed an issue where WebM files were treated as audio.
Fixed an issue that could cause authentication errors part way through a turn on slower network connections.
Permission prompts again describe the specific action being requested instead of showing a generic title.
Fixed an issue where subagents running side by side disappeared from the conversation when one of them reported progress.
Fixed an issue where an extension or automation could briefly show an unavailable error after being restarted.
A failing custom hook no longer ends the session; it is now reported as an error and the conversation continues.
Fixed the connected tool server configuration dialog overflowing its window.
Fixed the built-in workflow migration helper reporting wrong paths and moving the wrong files on Windows.
Fixed keyboard focus rings on sidebar items being clipped or rendering with square corners.
Generative UI renders HTML artifacts inline in chat, with support for KaTeX math and Chart.js and Plotly charts. This release also adds side-by-side terminal splits, the Darcula theme, and @path/to/file references inside AGENTS.md and custom rule files.
Improvements(10)
Added support for referencing and inlining external files directly using @path/to/file syntax within AGENTS.md and custom rule files.
Added support for rendering YAML frontmatter as a formatted metadata card in markdown files and artifacts.
Added support for KaTeX math formatting, Chart.js, and Plotly charts in generative UI widgets and artifact previews.
Added the ability to split terminals side-by-side in the terminal panel using a dedicated split button or keyboard shortcut (Ctrl+Shift+5 on Linux/Windows, Cmd+\ on Mac), complete with draggable divider resizing and tree indicators in the terminal list.
Added the Darcula theme preset and syntax highlighting color palette.
Added support for viewing specific page ranges (StartPage and EndPage) in multi-page documents (like PDFs) and specifying image resolution (MediaResolution) in the view_file tool.
Added support for hover preview card on skills chip in chat.
Added support for defining a rules: key in custom markdown agent frontmatter to bind specific rule files directly to an agent.
Added support for discovering custom skills, agents, and rules defined in skills.json, agents.json, and rules.json configuration files located within project subdirectories.
Several overall UX improvements for performance, accessibility, and project configurations.
Fixes(30)
Fixed an issue where slash commands unavailable to a selected custom agent were incorrectly shown in the command menu.
Fixed an issue where image attachments in the message composer were lost when switching between conversations.
Fixed an issue where Ogg audio and video files were assigned generic MIME types that could cause model errors.
Fixed an issue where custom color themes with transparency or alpha channels caused UI text to render in red.
Fixed an icon mismatch where the title bar overflow menu displayed a generic terminal icon instead of matching the tab icon in the side panel.
Fixed a memory leak where uncommitted conversation renders could leave agent state updates running in the background.
Fixed a memory leak where rapidly switching between conversations could leave background streaming connections open indefinitely.
Fixed an issue where clicking a comment link would not scroll to the comment if it was located below the visible fold in virtualized file and diff viewers.
Fixed an issue where project folder names with special characters or spaces were improperly decoded twice.
Fixed an issue where newly added files in the review pane displayed an inaccurate added line count due to trailing newlines.
Fixed an issue where users authenticated via Workforce Identity Federation (WIF / Enterprise sign-in) were logged out prematurely after one hour due to token refresh endpoint routing.
Fixed a z-index stacking conflict where inline file change panels overlapped the @-mention dropdown menu.
Fixed an issue where the side panel overflow menu incorrectly marked the last-clicked action with a checkmark.
Fixed an issue where form controls and scrollbars inside generative UI widgets did not adapt to the dark or light theme.
Fixed an issue where Enterprise users authenticating via Application Default Credentials could erroneously see an "Out of credits" error message.
Fixed cold-start token endpoint routing to prevent HTTP 429 resource exhaustion errors on launch, and cached account authentication status to eliminate redundant backend checks on user interactions.
Fixed an issue where long terminal commands in permission prompts caused the popup layout to overflow.
Fixed an issue where rapid consecutive agent status updates caused repeated notification chime sounds.
Fixed an issue where expanded tool steps and collapsible sections could collapse or remount when earlier steps updated during long conversations.
Fixed an issue where disabled menu items displayed an active pointer cursor instead of a disabled cursor.
Fixed an issue where previously configured command permission allow and deny rules were ignored after migrating to the updated permissions system.
Disabled the Git Push action in repositories without a configured remote, and added clear tooltips explaining why VCS actions are disabled.
Fixed an issue where pasting text into the integrated terminal using Ctrl+V failed unless clipboard read permissions were explicitly granted.
Fixed an issue where artifacts created in nested subdirectories could fail to appear in the artifacts panel.
Fixed an issue where signed-in users clicking "Manage" in Settings briefly saw an unauthenticated account overlay.
Fixed an issue where selecting text in an assistant response failed to show the "Quote" button if the mouse drag released below the message bubble.
Fixed authentication race conditions that could cause error loops when signing in again or leave stale account state after signing out.
Fixed an issue where a single corrupted or empty project file could cause the entire projects sidebar to appear empty.
Fixed an issue in the file viewer where selecting a single code token or word failed to display the comment button.
Fixed an issue where temporary network disruptions or DNS lookup timeouts during startup caused permanent account failure screens instead of automatically retrying.
Embedded Terminals, Git Version Control, and Audio Attachments
An embedded terminal and Git-native version control now live directly in the sidebar. This release also adds audio file attachments, interactive image commenting, richer previews for MCP tool execution, and core performance work.
Improvements(13)
Added an embedded terminal directly in the sidebar (Ctrl/Cmd + `), allowing you to run build commands, test suites, and scripts without switching applications.
Introduced Git version control in the Review pane to inspect working tree diffs, stage and unstage files, and commit changes directly within the app.
Improved Git responsiveness and reduced background CPU usage on large repositories and projects with submodules.
Added support for attaching audio files (MP3, WAV, M4A, AAC, OGG, FLAC, OPUS) up to 20MB via file picker or drag-and-drop.
You can now drag to select regions and leave comments directly on any image in the file viewer, with smooth scrolling for tall images.
Model Context Protocol (MCP) and custom tool execution steps now display structured headers, expandable argument blocks, and formatted result previews.
Redesigned pending review comments in the chat input with unified attachment badges, file-grouped hover previews, and quick modal access for queued messages.
Improved sidebar smoothness and message streaming responsiveness during active generation.
URL artifact cards now open supported links (such as local dev servers and Google Docs, Sheets, Slides, and Drive) directly in an in-app preview pane.
Improved folder picker navigation with live directory suggestions, Windows path support, and inline new folder creation.
File search in the @-mention menu is now typo-tolerant, making it easier to reference files with approximate names.
Markdown file previews and artifacts now cleanly format frontmatter metadata as a structured key-value summary.
Double-clicking an option in single-select questions and confirmation prompts now immediately selects and submits the answer.
Fixes(8)
Fixed an issue where the slash command menu and input suggestions displayed commands unsupported by the active custom agent.
Fixed an issue where goal progress and summary sections were missing from the chat interface during /goal execution.
Fixed an issue where unsaved edits to artifact comments could be lost when switching tabs or moving the cursor away.
Fixed an issue where comments on image files were missing their cropped preview and comment text in the comments modal.
Fixed an issue where image attachments in the chat composer were lost when switching between conversations.
Fixed an issue in project settings where selecting "Inherit General" for Sandbox Mode failed to save.
Fixed an issue where quoted text and context items in question headers appeared as raw markup instead of interactive pills.
Fixed an issue during business login where license lookup errors would leave users on an error screen instead of prompting for a Google Cloud project ID.
Remote Control, faster project switching, and more
This update introduces Remote Control to drive and monitor agent sessions on your local device from any browser, alongside core performance gains, improved sidebar navigation, and cleaner error recovery.
Improvements(11)
Introduced Antigravity Remote Control, allowing you to securely connect to and drive agent sessions on your local device from any browser, with proactive push notifications when agents complete tasks or require input.
Improved performance across the app: faster startup and project switching, faster loading of conversations with long histories, less lag during active sessions, and smoother opening and collapsing of the sidebar and side panels.
Added the ability to collapse top-level sidebar sections. The ungrouped list now shows up to 30 conversations, and conversation actions appear instantly on hover.
Notification toasts now appear in the bottom-right corner and collapse into a compact stack that expands on hover, so they no longer cover the screen.
Hovering over a file's comment count pill now previews all comments on that file alongside their line numbers, and clicking a comment opens the referenced file, diff, or artifact with that comment focused.
Refreshed the Customizations settings screen with collapsible sections and item counts for skills, rules, plugins, and custom agents, consistent back navigation across sub-pages, and clearer card layouts and status indicators for MCP server configurations.
Clicking Retry on an agent error card now resumes the task directly, without adding an artificial "Continue" message to your conversation.
Custom agents can now reuse the skills, rules, and subagents you already have, using a single inheritCustomizations setting in the agent's frontmatter.
Added support for uploading and attaching WebP images in chat.
Added syntax highlighting for PowerShell (.ps1, .psm1, .psd1, and pwsh/ps1 code fences) and Zsh (.zsh, zsh) in chat and the file viewer.
Refreshed the Google Docs, Sheets, Slides, and Drive icons to the 2026 Workspace logos, and URL artifact cards for those products now show the product icon instead of a generic link glyph.
Fixes(15)
Fixed an issue where Find (Ctrl+F / Cmd+F) in the Review pane missed matches that were outside the visible area or hidden inside collapsed files and folded line blocks. Search now scans all open files and automatically expands and scrolls to each match.
Fixed an error that could stop you from rewinding a long conversation.
Fixed an issue where a conversation could get stuck showing as still running and never return to idle, most often after pressing Stop.
Large binary attachments returned by connected MCP tools — images, PDFs, and audio — are now saved to a file and referenced in the conversation instead of being pasted inline, where they could previously overflow the conversation and break the turn.
Fixed an issue where one invalid or unsupported MCP server configuration would prevent all other configured MCP servers from loading.
Fixed an issue on Windows where the Proceed button failed to appear on artifact review cards and plan editor headers.
Fixed an issue where titles for long-running conversations could unexpectedly change to an older message.
Fixed an issue where pinning, archiving, or restoring a conversation that was no longer available on disk appeared to succeed. The action is now reverted and an explanatory notification is shown.
Fixed an issue where newly created conversations in the sidebar temporarily displayed raw link text instead of clean @-mention names.
Added visible error notifications when file attachments fail because of file size limits, attachment counts, or unsupported formats.
Fixed an issue where confirmation dialogs (such as reverting actions or deleting conversations) focused the close button by default, so pressing Enter dismissed the dialog instead of confirming it.
Fixed an issue where updated images in chat could keep showing stale, cached versions; they now refresh on reload.
Fixed an issue where comments added to markdown files or artifacts in preview mode attached to the first line instead of the selected text.
Fixed an issue where images uploaded or pasted into the chat composer alongside a comment on an artifact were left out of the message sent to the model.
Fixed an issue in the Review pane where dragging file header pills into the chat composer did not correctly reference files beyond the first.
Persistent Sidebar Folders, Crash Fixes, and UI Enhancements
Antigravity improves workspace organization by persisting your sidebar folder states. This release also ensures reliable file and artifact previews.
Improvements(1)
The sidebar now remembers which folders you have collapsed or expanded, including when you use the 'Collapse All' action, keeping your Projects list organized across sessions.
Fixes(6)
Fixed an issue that prevented submitting or dismissing the feedback form on the welcome screen.
Fixed frequent application crashes when opening conversations that contain large command outputs. The file viewer now limits the size of files it will display, and large changes in the history are limited to ensure stability.
Fixed a styling issue where inline code in chat messages and the file viewer did not display in the correct colors.
Fixed a layout bug that caused multi-line commands to display without line breaks in the activity history.
Fixed an issue where the application header and top menu bar failed to display on Linux.
Fixed an issue where empty section headers (such as "Terminal") appeared in Settings when all contained settings were disabled.
Reliability, Performance Gains, and User Experience Polish
Antigravity 2.7.1 introduces visual side-by-side previews for image diffs alongside comprehensive screen-reader and keyboard accessibility enhancements. This update also delivers key performance boosts, bug fixes, and UX refinements across chat interactions and project navigation.
Improvements(11)
Improved performance of the conversation sidebar, especially when managing a large number of projects and conversations.
Improved the performance of the conversation sidebar, reducing lag when switching between conversations.
Added the ability to delete projects from the Project Settings page. Deleting a project permanently removes it along with all associated active and archived conversations.
Added side-by-side previews for image files (including SVGs) when comparing file changes.
Added a 'Last Prompt' sorting option to the display options menu, allowing you to sort your conversation history by the time of your last message.
Improved the slash-command menu with smarter search ranking and highlighting, making it easier to find and select commands as you type.
Adjusted the appearance of the thumbs-up and thumbs-down feedback buttons on messages so they no longer look disabled.
The conversation hover card now displays the project name next to a folder icon, showing where the conversation is saved.
You can now use the keyboard to navigate to action buttons and menus that appear when hovering over list items (such as pinning or deleting conversations).
Improved sidebar organization with new conversation grouping options. You can now group conversations by project or status while keeping standalone conversations separate, or view all conversations in a single combined list.
Improved the design and organization of the display options menu in the projects sidebar and history pages. Additionally, menu checkboxes have been updated to display clearer selection states.
Fixes(20)
Ensured that local web servers started by the agent are reachable from your host machine.
Fixed a visual issue where submenu items (such as the 'Copy' options) were vertically misaligned when opened.
Fixed an issue where the chat input box lost focus after switching conversations using keyboard shortcuts.
Fixed an issue where scrolling inside the file path dropdown would also scroll the main file path bar in the background.
Fixed an issue where line numbers were incorrectly displayed on file pills for images, audio, and video files in the chat.
Fixed an issue on mobile devices where the keyboard remained open after sending a message.
Fixed an issue where selected checkbox options in menus changed text color, ensuring consistent styling across all menu items.
Fixed an issue where the Review tab would get stuck loading when trying to display binary files or files that failed to load, replacing the spinner with an informative message.
Fixed an issue where SVG images without specified dimensions would appear blank or collapsed when opened in the media viewer.
Fixed an issue where renaming a conversation from the header menu did not work on mobile web browsers.
Fixed an issue where SVG files failed to display and showed errors; they now render correctly as images in the file viewer.
Fixed an issue where folders added to the prompt displayed file icons instead of folder icons.
Fixed a bug where searching for text inside a file diff would fail to return matches that were scrolled off-screen.
Fixed an issue where searching for text within markdown previews returned no results.
Fixed a bug in the conversation display options menu where the subtitle setting could appear unselected. Conversation subtitles now display folder and branch icons next to project and branch names for improved visual clarity.
Fixed the project selector in the input box to ensure you can see and select all your projects when starting a new conversation.
Fixed an issue where search matches were not highlighted when searching within a file.
Fixed an issue where the highlighted option in the @-mention menu could scroll out of view when navigating with the arrow keys.
Fixed an issue where conversation titles in the sidebar would temporarily revert to "Untitled Conversation" while updates were loading.
Fixed an issue where project artifacts would fail to load if the conversation was not active in the sidebar.
Faster long conversations, more reliable hooks and subagents
Conversations with long histories now open faster, improved custom hooks and subagents behavior, and administrator policies for connected tool servers are applied correctly — plus a broad set of fixes across chat, comments and the file viewer.
Improvements(9)
Improved the ask questions panel: clearer option selection, and continue now only activates once you've answered the current question.
Hover details on the small icon chips in chat now appear almost immediately instead of after a noticeable pause.
Images and videos in chat now open in a preview window when clicked, with the option to open them in a tab instead.
Hook configurations that could never run are now rejected at load time with a clear error instead of being silently ignored.
Projects in the command palette now use the same folder icon as the sidebar, instead of a mismatched briefcase.
Opening long conversations is now dramatically faster.
You can now copy an image to your clipboard from an artifact or the file viewer, using the menu or right-click.
The conversation list now loads far faster for users with a large conversation history.
Settings descriptions now use consistent wording throughout, and two typos in the shortcut and host address settings were fixed.
Fixes(21)
Slash commands now work when a newline or tab separates the command from its arguments.
Conversations started by background automations no longer disappear from the sidebar list; they now appear under the scheduled-tasks filter.
Steps waiting on your input now show the icon for the tool involved, including in the permission prompt, instead of a generic bell.
Fixed the comment popup following your cursor while dragging, and whole-file comments landing on the first line with no highlight.
Agents no longer hang forever when a stop hook keeps blocking the end of a turn; the turn now finishes after repeated blocks.
Conversations that had already finished no longer stay stuck showing as running after you quit and reopen the app.
Fixed hovering a message revealing the Stop button on every running subagent card; now only the card you hover shows it.
The Review Changes button's keyboard focus outline now wraps the entire button instead of only part of it.
Hooks that call a model now stop at their configured timeout with a clear error, instead of stalling the agent indefinitely.
Pasting plain multi-line text into the chat box no longer adds an extra blank line between each line.
Fixed pinning and unpinning a conversation often failing on the first click and snapping back, forcing you to click repeatedly.
Commands still running no longer display a nonsensical negative elapsed time when their start time is missing.
Fixed administrator policies for connected tool servers being skipped at startup, which could let restricted servers start without their required settings.
Fixed agents losing access to their built-in helpers, so research, browser and teamwork assistants work again.
Fixed the wrong keyboard shortcut being displayed for toggling the side panel.
Fixed the built-in browser inspection tool being wrongly blocked when your organization restricts which tool servers can run.
Stopping a subagent now also stops everything it spawned, including nested subagents and their background tasks, instead of leaving them running.
Fixed long answers and previews periodically jumping back to the top and losing your scroll position while you read.
Custom hooks you define now run at the end of a turn instead of being skipped before they could fire.
Fixed a crash that happened when you selected a region of an image and submitted a comment on it.
Fixed a stray outline around the pending-step badge in the conversation list under some themes.
Preview tabs, MCP timeouts, file attachments, and performance fixes
Added preview tabs, attachment support for .json, .md, and .csv files, MCP timeouts, keyboard shortcuts, and various bug fixes and improvements.
Improvements(17)
Added a keyboard shortcut (Cmd+L / Ctrl+L) to quickly quote selected text into the chat input.
Added support for attaching .json and .md files to the input box with distinct visual badges.
Added preview (temporary) tabs to the auxiliary pane, allowing files to open in a temporary slot that is reused until promoted to a permanent tab (e.g., via double-click).
Added a timeout for MCP server connections and tool calls to prevent the agent from hanging indefinitely.
Added an 'Only Unread' filter option to the Conversation History dropdown.
Added support for uploading and attaching CSV (.csv) files in chat as plain text.
Added a 'Download Diagnostics' command to the command palette to download a diagnostics package (containing logs and agent state) for troubleshooting.
Added tooltips to file links in markdown views, showing their full path on hover.
Refined command approval rules to reduce redundant permission prompts for everyday tools.
Improved the speed of conversation switching by optimizing trajectory summary list selectors.
Improved authentication error messages to show the specific failure reason.
Draft comments are now automatically saved and preserved across sessions.
Improved interface responsiveness by preventing unnecessary re-renders when toggling sidebars or auxiliary panes.
Improved accessibility in the Settings modal with better heading hierarchy and keyboard focus management.
Updated the default settings screen name from 'Agent' to 'General' and updated copy for inherited settings ('Use Global' to 'Inherit General').
Restored the default 5-second auto-dismiss behavior for in-app notifications.
Optimized scroll performance in conversation lists and panels, especially during long chats.
Fixes(10)
Fixed an issue where configuring duplicate tool names in customizations (such as MCP servers) caused the agent to fail to initialize.
Fixed a potential crash in the image generation tool.
Fixed a bug where terminal commands with quoted arguments (e.g., arguments containing pipes or special characters) were incorrectly parsed, which could cause permission checks to fail or behave unexpectedly.
Fixed an issue where queued agent messages could be delivered and displayed out of order on Windows clients.
Fixed an issue where live-streamed subagent steps would display as generic text instead of rich cards.
Fixed an issue where inline markdown diff blocks could occasionally render as empty boxes.
Fixed a bug where the Settings dialog would close unexpectedly when clicking outside an open dropdown.
Fixed an issue where the UI would show an infinite loading spinner instead of falling back to agent edits when version control features were unavailable.
Added view options (grouping, sorting, filtering) and fixed search filtering on the Conversation History page.
Fixed an issue where the app doesn't request microphone permission when enabling "Record Audio" for the first time.
Queued messages, plain text files support, and stability improvements
Added support for queued messages, settings to configure message execution, a send now option, plain text file attachments, and various bug fixes and improvements.
Improvements(6)
Added support for queued messages, including settings to configure execution behavior, a "Send Now" option, and an improved input card layout.
Added support for attaching and rendering plain text (.txt) files in conversations and optimized comment button alignment in diff views.
Streamlined startup performance by optimizing file watching and the application initialization sequence.
Refined command permission prompting with strict literal matching, relaxed redirection rules, and sandbox configuration improvements.
Enhanced reliability with automatic retries on backend overload errors and resolved hangs in helper agents requesting feedback.
Updated the default UI theme to respect system preferences instead of defaulting to dark mode.
Fixes(5)
Fixed find-in-page (Cmd+F) search to correctly locate and highlight matches in virtualized file views.
Improved side questions (/btw) with persistence across conversation switches, display of the agent's thinking process, and a fix for background retries when closed.
Stopped background tasks from continuing to run after a conversation is archived.
Fixed an issue where diffs for files with more than 1000 lines rendered incorrectly.
Fixed an issue where the agent could hang indefinitely after subagents finished executing.
Antigravity Guide, audio support, search improvements, and performance fixes
New built-in Antigravity Guide skill, audio file rendering, improved substring file search, performance optimizations, and critical bug fixes.
Improvements(19)
Added a new built-in Antigravity Guide skill to provide helpful guidance when users ask about Antigravity.
Added syntax highlighting for C++, Python, and Protobuf files in markdown code and diff blocks.
Added hover tooltips to file pills in the chat and workspace views to display absolute paths and clarify workspace locations.
Improved readability of context pills by truncating long labels and adding hover tooltips to view the full content.
Enabled automatic saving of refreshed OAuth tokens to the OS keyring to reduce authentication prompts.
Added support for rendering and playing audio files (.mp3, .wav, .ogg, .m4a) within the sidebar file viewer and artifact viewer.
Added a new 'Conversation Width' setting under Appearance settings, allowing you to configure the maximum width of the conversation panel (Default, Narrow, or Wide).
When creating a new project, Antigravity 2.0 will now notify you if a project with the same folders already exists, allowing you to quickly navigate to it.
Added a tooltip to show the full task summary when hovering over items in the Running Items panel.
Improved the model selector dropdown to dynamically resize and prevent truncation of long model names.
Updated the permissions request dialog to display a clear description of the action or command being run.
Improved response time and reliability of account status checks when Terms of Service violations are encountered.
Improved UI responsiveness and eliminated lag when resizing the sidebar layout.
Aligned theme colors, font sizes, text size, and line height in the chat input area to match surrounding UI density and ensure consistent rendering.
Improved usability for touch and stylus users by increasing the size of various UI hit targets, including buttons, pills, and toolbars.
Adjusted the diff viewer layout to provide more horizontal space for code contents.
Added a brief delay before displaying tooltips for long conversation titles in the sidebar.
Optimized file watching notifications to prevent visual flickering and improve performance during high-frequency file writes.
Updated icons for built-in slash commands and redesigned mentions/slash commands menu alignment in the chat input box.
Fixes(17)
Fixed an issue with broken text selection anchoring when adding comments to files or artifacts.
Fixed an issue where workspace context was lost when navigating to the history page or other views.
Fixed performance issues, rendering loops, and lag when navigating folders using the file path breadcrumbs or the file tree.
Fixed erratic behavior and lag in slash command autocomplete during network instability.
Improved file search in your workspace by matching substrings instead of requiring exact prefix matches.
Fixed an issue where file search within workspaces could fail with a 'No such file or directory' error.
Fixed an issue where allowing commands with special characters (like dots or environment variables) could trigger an infinite permission prompt loop.
Improved command execution permission matching to support prefix-matching for build/test commands and correct handling of quoted arguments.
Fixed issues where in-progress prompts and local history edits were discarded during navigation or permission checks.
Fixed an issue where custom theme settings and color modes were not correctly applied on startup, including rejecting hex codes with a leading '#' symbol.
Fixed startup issues and recurring User Account Control (UAC) prompts on Windows by registering a scheduled task and resolving access control errors on system PATH directories.
Fixed crashes and synchronization issues when deleting persistent state across tabs.
Improved accessibility and usability of the artifact review dialog, including contrast, tooltips, and error messaging.
Fixed an issue where the agent was blocked from reading builtin customizations and skills due to sandbox restrictions.
Fixed a startup race condition that could cause duplicate project folders to be created.
Fixed a deadlock issue that could cause subagents to hang during execution.
Fixed a potential crash when calculating token usage.
Quota screen redesign, PDF attachment support, new /btw slash command, and other bug fixes
Improvements(7)
Quota screen redesign: clearer, unambiguous view into “used” versus “remaining” credits in the Models tab of the Settings screen.
Ask side questions via /btw: while in a conversation, type ‘/btw’ in the input and select the ‘btw’ option in the menu to write a message to an ephemeral, single-response agent that has the context of your current conversation.
Simple conversation search functionality: cmd/ctrl+F to search for visible text in the conversation view.
Support for PDF attachments to messages to Gemini models: drag and drop PDFs from your filesystem or add PDFs using the media option in the Add Context menu button in the input box.
Breadcrumbs in file viewer pane header: quality-of-life UI to more easily see and navigate directories in your repo.
Nested subagents in the overview pane: see all nested subagents belonging to the main conversation instead of only the subagents that are one level deep.
Minor improvements to Projects UX: you can now specify a project name during the creation flow and sort conversations in the left sidebar by worktree.
Fixes(4)
Improved LaTeX support: the agent has a better understanding and awareness of its ability to output LaTeX-rendered math text.
Improved MCP server stability: increased resilience to unresponsive MCP servers and improved browser agent self-troubleshooting abilities when using the Chrome DevTools MCP server.
MCP server schema compatibility: the mcp_config.json schema now accepts url in addition to serverUrl as a field.
New entries in the sensitive paths list: .vscode and .cache are now recognized as sensitive paths that require explicit user confirmation before the agent can access them.
Model reasoning effort level controls, unconfigured plugin handling, and terminal rendering and selection fixes
Introduces reasoning effort level controls selectable with --effort or /effort, adjusts unconfigured plugin loading behavior, restores ASCII fallbacks for Mermaid diagrams and LaTeX in WezTerm and VS Code terminals, improves artifact viewer copy interactions, and fixes project custom agent discovery across workspaces and headless runs.
Improvements(2)
Improved reasoning effort level for models with different support, selectable with --effort or from the effort gauge in /effort and /model.
Changed plugins placed directly in ~/.gemini/config/plugins so that a plugin whose MCP server needs configuration variables now starts disabled until you enable it, matching plugins installed through /plugin.
Fixes(3)
Fixed Mermaid diagrams and LaTeX rendering as rows of garbled placeholder characters in WezTerm and the VS Code integrated terminal; these terminals are no longer assumed to support Kitty graphics and now show the ASCII fallback.
Fixed copying text from the artifact viewer when Copy on Select is turned off in /config: the viewer no longer captures the mouse, so your terminal's own selection and copy shortcut work, and Ctrl+C goes back to interrupting or exiting.
Fixed project custom agents in .agents/agents/ not being found or selectable in workspaces created in the Desktop App, in already-trusted workspaces, under execution with --agent, in headless (-p / --prompt) runs, and in the /agents panel.
Medium verbosity mode, artifact viewer background rendering, sandbox improvements, and error handling fixes
Introduces medium verbosity mode to group related tool calls into concise summaries, pre-renders Mermaid diagrams and LaTeX in the artifact viewer for Kitty terminals, enhances terminal sandbox behavior and directory configuration scanning, and fixes headless run exit codes and subagent worktree artifact handling.
Improvements(6)
Added medium verbosity mode to /config, between high and low, which groups related tool calls and thoughts into concise summaries (such as Explored N files) while keeping commands and responses visible; the Verbosity setting and each option now describe what they show.
Improved Mermaid diagram and LaTeX rendering in the artifact viewer on Kitty-compatible terminals: artifacts are pre-rendered in the background so diagrams appear as soon as the viewer opens, images are uploaded once at a smaller size, and zooming with Ctrl+= / Ctrl+- no longer flickers between the old and new sizes.
Improved the artifact viewer footer by combining the separate top and bottom hints into one g/G top/bottom entry, labeling m with the view it switches to (diagram, LaTeX, ASCII, or raw), and fixing hints that contain arrow symbols wrapping onto a new line too early.
Improved terminal sandbox behavior so the agent asks to bypass the sandbox less often: it now knows sandboxed commands can read and write its own artifact and scratch directories, and it tries the sandbox first even after an earlier command needed a bypass.
Changed the step title of commands that exit with a non-zero code from Errored to Failed.
Changed how directory entries in skills.json, rules.json, agents.json, and plugins.json are scanned: an entry now loads only the items directly inside the directory, the same as a .agents/skills/ folder, instead of recursively loading everything beneath it; to load a nested item, name it in include_only, for example {"path": "shared_skills", "include_only": ["category/my-skill"]}.
Fixes(5)
Fixed pressing Esc while the suggestions dropdown is open interrupting the agent's turn; it now closes the dropdown, and a second Esc interrupts.
Fixed Mermaid diagrams and LaTeX leaving a blank gap inside tmux or GNU screen when the outer terminal supports Kitty graphics; the CLI now falls back to ASCII diagrams unless images actually reach the terminal, and CLI_GRAPHICS=kitty still forces image mode for setups with image passthrough configured.
Fixed headless (-p / --prompt) runs that streamed part of a response and then ended on a model or agent error exiting with code 0; they now exit with code 3 and print the AGY_ERROR line, and JSON error output includes the partial response, while multi-turn stream-json sessions still warn and continue.
Fixed machines set up with the old Remote Control installer script crash-looping after an upgrade; when the CLI is launched by that deprecated background service it now unregisters the service and points to remote-control start instead of restarting repeatedly.
Fixed files that subagents write inside their own Git worktree being treated as artifacts, which demanded artifact metadata and left .metadata.json files in the worktree; subagent worktrees now live under a worktrees/ directory in the app data directory.
Direct subagent messaging syntax, Vim numeric count multipliers, headless lifecycle handling, and concurrency fixes
Introduces direct @ prompt syntax with interactive autocomplete, adds Vim count multipliers across Normal and Visual modes, enhances headless execution lifecycle and task timeout handling, and fixes file-lock concurrency issues in conversation history databases along with streaming response stability fixes.
Improvements(6)
Added @ prompt syntax to send a message directly to a subagent conversation, with autocomplete listing running and completed subagents.
Added Vim numeric count multipliers so counts apply to operators, motions, and actions in Normal and Visual modes, including 3dw, 2d3w, 3dd, 3x, 3rX, 3p, 3u, [count]G/gg/$, and counted text objects such as 2di(.
Improved GEMINI_API_KEY sessions to reduce behavior discrepancies with the non-API-key sign-in path.
Improved the artifact viewer to show a Left/Right pan hint in the footer when a wide diagram overflows the window.
Improved /rewind to show a relative timestamp for each step and to focus the most recent step when the panel opens.
Improved command allow-listing suggestions to recognize jj config and jj op subcommands when offering to always allow a command.
Fixes(9)
Fixed headless (-p / --prompt) runs leaving daemon background processes running after exit, which could hang scripts reading the CLI's output until end-of-file; daemon processes now terminate when the run ends.
Fixed headless (-p / --prompt) runs cancelling still-running background tasks about 5 seconds after the agent went idle; runs now wait for background tasks until the --print-timeout deadline, up to a 30-minute cap.
Fixed a conversation-history database corruption risk where checking the database file for write access could silently drop file locks held by concurrent CLI processes on the same file.
Fixed context compaction failing when the tool configuration used for compaction checkpoints was rejected in certain scenarios.
Fixed a backend crash when a streamed model response chunk arrived without its response envelope, which terminated the session with connection errors.
Fixed markdown table column alignment when a table cell contains file links that wrap across lines.
Fixed markdown file links to code symbols dropping their display text and rendering the raw path instead.
Fixed incomplete enterprise sign-ins (for example closing the browser window before finishing license or project selection) leaving behind a stuck partial credential; such credentials are now cleared automatically so sign-in can be retried cleanly.
Fixed the browser companion page title to read Antigravity CLI instead of Antigravity Cli.
Proportional context compaction budgets, full-window sizing, custom model media support, and leak fixes
Improves context compaction by spreading user-request budgets proportionally across all prompts and sizing truncation ceilings from the model's full context window, enables default PDF and audio support for custom models, routes Ctrl+G editor launches to a real terminal stdout, caps voice dictation recordings to avoid deadline failures, and resolves resource and memory leaks across conversations and background pollers.
Improvements(2)
Improved context compaction to spread its user-request budget across all captured prompts, so a long initial instruction is no longer truncated to a small fixed slice when the other prompts in the conversation are short.
Improved context compaction to size summary and truncation budgets from the model's full context window instead of the compaction trigger threshold, so compacted summaries and background-task lists are no longer prematurely cut short.
Fixes(9)
Fixed PDF and audio tool outputs and attachments failing with unsupported mime type errors on custom models configured in settings.json; custom models now accept PDFs by default and honor the modelFeatures media flags for images, video, PDF, and audio.
Fixed pressing Ctrl+G to edit the prompt in a full-screen terminal editor such as vim hanging with Vim: Warning: Output is not to a terminal; external editors now run against a real terminal stdout.
Fixed voice dictation sessions longer than 4 minutes failing with a deadline error and erasing the drafted transcription; recordings now automatically stop and finalize at 3 minutes 30 seconds.
Fixed the startup banner displaying a Google Cloud Project ID for consumer (non-enterprise) sign-ins.
Fixed a stack-overflow crash when loading or compacting conversations containing background-task, subagent-management, messaging, or scheduling steps.
Fixed forked conversations inheriting step output data from steps after the fork point, which could collide with new steps produced in the fork.
Fixed a background poller and timer leaking for every conversation session, slowly accumulating memory over long-running sessions.
Fixed canceled conversation and workspace creation requests continuing to provision resources in the background; aborted requests now stop immediately.
Fixed quitting the CLI taking around 5 extra seconds before the process exited; shutdown now cancels open streaming connections immediately instead of waiting for a forced timeout.
Inline Kitty graphics for math and Mermaid diagrams, interactive ask_question navigation, dedicated rule token budgets, and 30s retry backoff
Introduces inline Kitty graphics rendering for LaTeX math and Mermaid diagrams in the artifact detail viewer with ASCII fallback and horizontal panning, interactive question navigation with write-in previews in ask_question, dedicated 20,000-token rule budgets to prevent customization eviction, and caps model retry backoff at 30 seconds, alongside plugin lifecycle hardening, Bubble Tea v2.0.9 upgrades, and headless background task logging.
Improvements(7)
Added inline Kitty graphics rendering for LaTeX math equations ($$...$$ and fenced math, latex, and tex blocks) and Mermaid flowcharts and sequence diagrams in the artifact detail viewer, with Unicode ASCII fallback, Left/Right arrow horizontal panning for wide diagrams, and m to cycle between Image, ASCII, and Raw views.
Improved the ask_question interactive prompt with Left/Right arrow navigation between questions, inline previews of confirmed write-in answers, check marks on answered options, and single-step submission on the final question.
Improved the CLI startup banner for enterprise accounts to display the active Google Cloud Project ID beneath the signed-in account and plan tier.
Improved model API retry responsiveness by capping per-attempt retry backoff at 30 seconds instead of waiting up to 4 minutes between attempts.
Improved customization token budgeting by giving user and workspace rules a dedicated 20,000-token budget (cutting oversized rules on newline boundaries and listing over-budget rules by path and description) so large rule sets no longer evict skills, workflows, subagents, or MCP tools.
Improved the /usage panel by removing the duplicate remaining-quota percentage line beneath each progress bar.
Improved the default agent and subagent toolset by retiring the legacy find_by_name, grep_search, and list_dir tools from the default baseline while keeping them available to custom agents that explicitly list them in tools.
Fixes(6)
Fixed plugin skill slash commands being double-prefixed (/::) when a skill's frontmatter name already includes the plugin prefix, or being shadowed when multiple plugins define skills with the same short name.
Fixed plugin upgrades leaving old MCP server and sidecar processes running from deleted install directories, and pruned superseded plugin versions and staging directories from the marketplace cache on startup.
Fixed disabled plugins disappearing from the /plugin list after enabled-only customization queries, and hardened plugin ID path validation on uninstall.
Fixed terminal rendering and Kitty keyboard protocol stack handling on exit and screen clear by upgrading Bubble Tea to v2.0.9.
Fixed headless (-p / --prompt) runs occasionally skipping the background-task waiting notice, logged background SDK tool progress to task log files, and reduced memory usage by cloning truncated command output previews.
Fixed starting the CLI unpinning conversations or marking them unread in the desktop app.
Session-scoped Remote Control, unlimited headless run timeouts, structured error reporting, and artifact text selection
Introduces session-scoped Remote Control via --remote-control and /remote-control, lifts the default 5-minute timeout on headless runs, adds structured AGY_ERROR output on stderr, improves word-wise navigation in the prompt editor, and fixes Remote Control workspace permission inheritance, oversized file edit storage limits, full-screen artifact text selection, queued /model switching, and /rewind snapshot resolution.
Improvements(4)
Added Remote Control (start a connection via --remote-control startup flag or /remote-control slash command) to create a session-scoped remote connection for following and controlling your active terminal session from another device. Typing /remote-control off or closing the session automatically tears down the tunnel and unregisters the device from the active Remote Control session list.
Changed the default timeout for headless (-p / --prompt) runs from 5 minutes to unlimited so long-running agent turns run until the response completes unless --print-timeout is passed explicitly, and enabled daemon background commands in headless GEMINI_API_KEY sessions so background servers stay running after the turn finishes.
Improved headless (-p / --prompt) error reporting when a turn terminates on an agent or model API failure: the CLI now prints a structured AGY_ERROR: {...} JSON line on stderr with canonical status, HTTP or gRPC error code, retryability, and error ID (including HTTP status mapping for GEMINI_API_KEY SDK errors) and exits with code 3 instead of 1.
Improved word-wise cursor movement (Alt+F, Alt+B, and Ctrl/Alt+Arrow keys) and word deletion (Ctrl+W, Alt+Backspace, and Alt+D) in the prompt editor to stop at punctuation boundaries instead of only whitespace, making it easy to step through or delete individual segments of file paths, URLs, and flags.
Fixes(6)
Fixed turns triggered from a connected Remote Control session or running across secondary workspaces executing without the CLI session's active permission mode, cycle mode, and non-workspace file access grants.
Fixed artifacts in the Remote Control companion rendering as plain .md file links instead of rich artifact cards when the CLI's application data directory differs from the default bundle name.
Fixed text selection in the full-screen artifact viewer capturing the line-number gutter and trailing padding; dragging across the viewer now highlights and copies only the document text, with Ctrl+C to re-copy an active selection, Esc to clear it, and Shift+drag for native terminal selection.
Fixed submitting /model while a turn is already running switching the active model immediately in the middle of the in-flight turn; the one-shot model switch now waits until the queued prompt begins executing so the running turn and any earlier queued prompts finish on the session's original model.
Fixed file edits with diffs larger than 1 MiB exceeding the conversation storage limit and force-clearing the session; oversized edits now preserve their line-change statistics while omitting the raw diff body from stored conversation history and skipping the elided diff during /rewind reverts.
Fixed /rewind conversation reverts and forks sweeping backward to an older workspace snapshot when recent snapshot commits were skipped, which could silently overwrite kept work; snapshot lookup now sweeps forward from the target step and falls back to step-by-step revert replay when no later snapshot exists.
Introduces automatic subagent roster and instruction injection for custom Markdown agents, preserves explicit background task names across notifications and lists, cleanly clears invalidated auth credentials, and fixes interrupted command exit codes, remote cancellation indicators, artifact viewer window resizing, and frozen thinking frames.
Improvements(2)
Added automatic subagent guidance for custom agents defined in Markdown: agents that list invoke_subagent in their tools now receive the roster of available subagents and usage instructions in their system prompt, so they can actually delegate to the subagents they declared.
Improved background task naming so a task keeps the name it was given when sent to the background; completion notifications and task lists now show that name instead of a generic auto-derived one.
Fixes(5)
Fixed stale credentials lingering after the sign-in server definitively rejects them (for example a revoked or invalidated enterprise account); the CLI now signs you out and clears the stale tokens so you can sign in again cleanly, while temporary network failures, server errors, and rate limits never cost you your session.
Fixed commands killed by an outside signal (for example when the whole session shuts down mid-run) being recorded as successful runs with exit code 0; such commands now finish as canceled while keeping the output collected before the interruption.
Fixed the Interrupted hint only appearing when Esc was pressed locally in the terminal; it now appears whenever the current turn is actually cancelled, including when it is stopped from a connected remote session.
Fixed the artifact viewer leaving markdown text stuck at its original width after a terminal resize; content now re-wraps to the new window size.
Fixed in-progress thinking frames occasionally freezing permanently in the conversation history (for example showing Thinking... (31s) forever), including after a terminal resize or session reload.
Live skill reloading, interactive model search, mid-session thought signature fixes, and self-correcting tool validation
Adds /skills reload for live skill discovery, introduces fuzzy search and autocomplete for /model, improves OAuth scope error guidance, and resolves mid-conversation model switching errors with Gemini thinking models, search_web summarization failures, tool schema validation recovery, and daemon conversation resumption.
Improvements(3)
Added a /skills reload subcommand to asynchronously reload discovered skills and slash commands without restarting the session or blocking user input.
Added an inline argument autocomplete dropdown when typing /model and a live fuzzy search bar inside the interactive /model picker to filter models by name or ID.
Improved authentication error messages when OAuth tokens lack required scopes by including explicit instructions to run /logout and /login.
Fixes(10)
Fixed mid-conversation model switching to Gemini thinking models failing with thought_signature validation errors when prior turns were generated by non-Gemini models.
Fixed web search (search_web) failing with "no summary returned from GenerateContent" when the summarization model emits leading thought parts.
Fixed agent turns terminating prematurely with NO_TOOL_CALL when a tool call fails schema validation, allowing the agent loop to return the validation error and self-correct.
Copy support for side questions, native backend tool-choice constraints, and subagent MCP inheritance
Introduces /copy btw to copy side-question responses directly to the clipboard, translates tool-choice constraints to native backend function calling, and fixes plugin hook discovery, subagent MCP server inheritance, relative config path resolution, and premature proactive feedback prompts.
Improvements(2)
Added the /copy btw slash subcommand to copy the full text of the active /btw side-question response to the clipboard even when the response card is collapsed or scrolled, along with argument ghost hints when typing /copy.
Improved tool-calling fidelity by translating per-request tool-choice constraints (any, required, and named functions) into native backend function-calling configurations.
Fixes(3)
Fixed the /hooks command and hook inspection utilities omitting hooks bundled inside enabled plugins when listing active hooks.json configurations.
Fixed custom subagents created with enable_mcp_tools: true receiving an empty MCP server list instead of inheriting the parent agent's configured MCP servers, and fixed declarative subagent configurations failing to resolve relative config file paths (relative_path_to_config) against the parent agent's directory.
Fixed proactive feedback prompts triggering prematurely during the first few turns of a session or on transient, recovered tool errors.
Enhances startup migration diagnostics for unsandboxed permission rules, speeds up /resume with stale caches, automatically namespaces plugin MCP servers to avoid collision, resolves step-cache goroutine leaks, and fixes critical issues with Gemini API thinking block retention, oversized binary file loading, artifact reviews, and database cleanup.
Improvements(2)
Improved the startup warning for deprecated unsandboxed permission rules across CLI, shared, and project configuration files to list each affected file path, up to five offending rules, and step-by-step instructions for migrating them to command rules.
Improved /resume startup responsiveness when opening a large conversation history with a cold or stale summary cache.
Fixes(7)
Fixed MCP servers bundled inside plugins colliding with each other or with user-configured servers in mcp_config.json when they shared the same server name by automatically namespacing plugin MCP servers as _.
Fixed a memory leak where opening or scanning conversations left background step-cache eviction goroutines running for the rest of the session, significantly reducing memory usage after opening /resume or switching conversations.
Fixed view_file attempting to parse non-UTF-8 binary files as text or loading files larger than 100 MB into the model context; unsupported binary formats and oversized files are now rejected with a clear error before overflowing the context window.
Fixed Gemini API (GEMINI_API_KEY) sessions dropping model thinking blocks from prior turns on subsequent user messages and failing to propagate thought signatures on text and thought parts.
Fixed artifact review failing to trigger when a directory name above .gemini/ matched a skipped path component such as scratch, prevented conversation forks and snapshot reverts from copying internal .system_generated/subagents and .system_generated/worktrees directories, and cleaned up subagent metadata records when deleting a conversation.
Fixed deleted conversations being recreated as empty, schema-less SQLite database files when background queries reconnected after deletion.
Fixed conversations launched without a workspace folder inheriting workspace paths and customizations from other open sessions.
Custom agent component exclusions, resilient model API retry handling, and conversation continuation fixes
Introduces excludeDefaultComponents for custom agent Markdown definitions, adds automated exponential backoff retries for transient model API errors, improves MCP open object schema validation, and fixes issues with --continue workspace resolution, inline terminal screen flashes, Remote Control auth display, and sandbox status indicators.
Improvements(4)
Added support for excludeDefaultComponents: true in custom agent Markdown frontmatter, allowing custom agents to opt out of default prompt sections and built-in tools while preserving post-invocation hooks.
Improved model API error resilience and diagnostics: transient genai.APIError failures (502, 503, 504, per-minute 429 rate limits, and mid-stream interruptions) automatically retry in-process with exponential backoff while preserving completed tool call outputs, and unrecovered 503 and 429 responses surface clear user-facing error messages.
Improved MCP and provider tool schema validation to preserve open object schemas (such as {"type": "object"} or explicit additionalProperties: true) instead of rejecting undeclared arguments on schemas that allow them.
Improved per-turn responsiveness and reduced memory usage when opening large conversations.
Fixes(5)
Fixed --continue starting a brand-new conversation when launched from a subdirectory, after a crash, or while another session is open in the same workspace; it now falls back to the most recent non-empty conversation in the current workspace or its parent/child directories.
Fixed full-screen blank flashes in inline mode when content first pushes to scrollback, at the end of a turn, or when clearing the screen with Ctrl+L.
Fixed remote companion UIs connected to an interactive CLI session via Remote Control displaying an unauthenticated sign-in screen instead of the active session's signed-in state.
Fixed the status line reporting the terminal sandbox as disabled when the session was launched with the --sandbox command-line flag.
Fixed image zoom keys (Ctrl+= and Ctrl+-) and the footer zoom hint appearing in the artifact viewer when a Mermaid diagram is rendered in ASCII mode instead of as a Kitty image.
Background service registration with remote-control, expanded half-page scrolling, and content safety reporting
Introduces remote-control service manager subcommands to run the CLI as a persistent background daemon, broadens half-page scrolling support and keybindings, and resolves issues with content filter handling, scratch directory watchers, plugin MCP initialization, schema validation, and conversation resumption.
Improvements(2)
Added remote-control start, remote-control status, and remote-control stop subcommands to register the CLI with your operating system's service manager as a background daemon that persists across logouts and reboots, including --name for custom instance labels and --session to scope to the active login session.
Improved half-page scrolling across all views (altscreen mode, diff viewer, and detail panels where Ctrl+D no longer triggers exit prompts) and added Shift+Up and Shift+Down default keybindings (navigation.half_page_up and navigation.half_page_down).
Fixes(8)
Fixed prompts or model responses blocked by content safety filters failing with generic errors or empty turns by surfacing explicit content-filter stop reasons.
Fixed temporary files in scratch/ directories triggering recursive filesystem watchers and cluttering artifact reviews and checkpoints.
Fixed MCP servers bundled inside global plugins failing to initialize at CLI startup or failing to update running states when plugins are toggled.
Fixed schema validation errors for third-party MCP server tools whose input schemas omit additionalProperties.
Fixed redundant filesystem customization discovery walks running on every message when no slash commands were invoked.
Fixed spurious ERROR: logging before google.Init log lines appearing in cli.log and background daemon outputs.
Fixed macOS background service lifecycle issues, legacy script cleanup, and paid Google Cloud project license availability for Remote Control.
Fixed older conversations failing to load with unknown step type errors upon resumption.
Model API retry resilience, headless print optimizations, and detailed tool approval prompts
Improves resilience to transient model API errors with extended exponential backoff, accelerates startup and sign-in by reading cached credentials, streamlines headless execution and error reporting, enhances tool approval prompts with contextual action descriptions and reasons, and fixes issues across plugin reinstalls, plugin MCP working directories, subagent lifecycle states, and command memory leaks.
Improvements(9)
Improved resilience to transient model API errors by retrying errors such as 503 Unavailable with extended exponential backoff to prevent brief hiccups from aborting sessions.
Improved sign-in and startup speed by reading signed-in identity details from stored credentials instead of making network requests on every launch.
Improved headless (-p) runs to exit promptly upon delivering final answers, waiting for non-daemon background tasks and scheduled timers within --print-timeout while preserving running background daemons.
Improved failure reporting in headless (-p) runs by printing fatal errors to stderr with stable error: markers and adding truncation notifications.
Improved headless (-p) latency by eliminating up to 200 ms of idle delay per turn and skipping unused conversation title generation model calls.
Improved tool approval prompts to display specific action requests (e.g. Run this command?, Allow access to this URL?, or Allow calling this tool?), restricting command editing to commands and including a Reason: explanation when triggered by hooks or cross-project files.
Improved model selection auditability by logging alias resolutions, --effort variant mappings, and deprecated model replacements in cli.log.
Changed --print-timeout expiration behavior in headless mode to return partial output and exit successfully with a warning on stderr instead of raising a timeout failure.
Changed default URL fetching permissions to prompt for approval by default before reading external URLs unless access has been pre-granted.
Fixes(7)
Fixed plugin reinstalls leaving behind files deleted from the source by replacing managed plugin directories cleanly, preventing self-directory installation corruption, and cleaning up disabled state entries from config.json upon uninstall.
Fixed plugin-defined MCP servers resolving relative working directories incorrectly by scoping them directly to the plugin's root directory.
Fixed a startup race condition where subagents could fail to discover tools from MCP servers that were still initializing.
Fixed subagents occasionally appearing stuck in running states in large conversations, resolving delays delivering queued idle messages.
Fixed a memory leak where finished terminal command executions retained internal state throughout long-running sessions.
Fixed headless (-p) runs hanging indefinitely on interactive implementation-plan approval prompts by proceeding through plan review automatically in non-interactive mode.
Fixed sign-in failing with project access errors after switching from a business account with a selected project to a personal account by resetting cached project and license data.
Mid-conversation model execution, custom agent subagent dependencies, and MCP parameter validation fixes
Introduces /model for running one-off prompts with alternative models, adds conversation_title metadata to status line scripts, adds subagent dependency declarations in custom agent Markdown frontmatter, and delivers fixes for MCP argument schema validation, headless permission handling, Vim normal mode shortcuts, and cross-conversation artifact access prompts.
Improvements(7)
Added /model , which runs a single prompt on another model and then returns the session to the model it was using, so you can consult a different model mid-conversation without disturbing your saved default.
Added a conversation_title field to the JSON payload passed to custom status line and window title scripts, so they can show the active conversation's name and pick up renames immediately.
Added an agents list to custom agent Markdown frontmatter, letting an agent declare the subagents it depends on using the same workspace-relative, absolute, and agent-relative path rules as skills.
Improved /model to show a [model] argument hint as you type, so the command's arguments are discoverable without opening help.
Improved the /settings panel so each Verbosity option explains itself as you highlight it, describing whether tool calls, commands, and thoughts are shown in full or collapsed into a token metrics table.
Improved the sign-in and Google Cloud setup screens to use the CLI's standard styled key hints, which follow your color scheme and keybindings and drop the navigation hint when only one option is available.
Changed /model to report Model already set to when the session is already using that model, instead of reporting a switch that did not happen.
Fixes(6)
Fixed tool calls to MCP servers accepting arguments the server's own schema never declared, so an invented parameter is now rejected and corrected instead of being silently dropped.
Fixed headless runs with -p silently skipping tool actions they were not permitted to take, which now end with a notice naming the refused actions and report them as denied_actions in the JSON output.
Fixed ? in Vim Normal mode opening the shortcuts panel while the prompt already contained text, so character-search and replace operators such as f?, t?, and df? work; ? still opens shortcuts on an empty prompt.
Fixed /model with no argument doing nothing when run before the session had finished starting up, so it now opens the model picker.
Fixed print mode (-p) exiting before the session finished shutting down, which could drop the run's trailing conversation history before it reached disk.
Fixed a redundant approval prompt when the agent reads or writes another conversation's artifact files after you have already allowed access to files outside your workspace.
Adds half-page scrolling (Ctrl+D / Ctrl+U) to the artifact viewer, adds the pickerGrouping configuration setting for /resume, improves Mermaid flowchart ASCII rendering, updates unselected model reasoning defaults to medium, and resolves issues with worktree cleanup, SQLite WAL checkpointing on exit, and subagent approval prompts.
Improvements(5)
Added half-page scrolling with Ctrl+D and Ctrl+U to the artifact viewer.
Added the pickerGrouping setting to /config and settings.json to configure the default conversation list view in /resume (flat or grouped by workspace).
Improved terminal ASCII diagram rendering for Mermaid flowcharts in the artifact viewer and conversation.
Improved /logout execution time by short-circuiting token removal directly to file storage when keyring storage is bypassed or unreachable.
Changed unselected model families in the interactive /model picker to default to medium reasoning effort instead of low.
Fixes(4)
Fixed subagent tasks unexpectedly prompting for tool approval in always-proceed mode while viewing the subagent details panel.
Fixed orphaned Git worktrees accumulating on disk under .system_generated/worktrees when killing subagents or deleting conversations.
Fixed SQLite database WAL checkpointing on CLI exit so trailing session metadata updates are flushed to disk before shutdown.
Fixed customization discovery logging spurious errors when accessing temporary directories or paths outside the active workspace.
Adds an opt-in workspace-grouped view to the /resume picker with Ctrl+F toggling, adds Gemini 3.8 Flash to the model catalog for GEMINI_API_KEY users, updates Markdown-defined custom agents to inherit ambient skills, rules, and subagents by default, and delivers fixes for long MCP OAuth authorization codes, Windows skill path separators, /boost worker tools, and runtime nil pointer crashes.
Improvements(3)
Added an opt-in workspace-grouped view to the /resume conversation picker, allowing users to toggle between a flat list and conversations grouped by directory with Ctrl+F.
Added Gemini 3.8 Flash to the model catalog when connecting with a GEMINI_API_KEY.
Changed custom agents defined in Markdown to inherit ambient skills, rules, and subagents by default, matching the configuration of default agents.
Fixes(7)
Fixed MCP OAuth authentication failing when authorization servers return authorization codes longer than 1024 characters.
Fixed skill path matching and grouping in the /skills panel on Windows where host path separators (\) caused global and workspace skills to be misclassified.
Fixed relative path resolution in Markdown-defined custom agents so relative subagent and plugin paths resolve correctly against the agent definition's directory.
Fixed duplicate permission grants accumulating in configuration settings across session reloads and subagent invocations.
Fixed /boost failing at runtime by improving worker tool configurations.
Fixed a fatal nil pointer crash in the agent runtime caused by background summary updates attempting to read cached steps after trajectory closure.
MCP panel navigation improvements, multi-line comment support in mcp_config.json, and headless stream cleanup
Improves /mcp panel arrow key navigation and action cycling, adds support for comments and trailing commas in mcp_config.json, ensures headless CLI streams close cleanly on exit, and delivers fixes for duplicate agent listings, deleted working directory startup, and /btw side questions with active goals.
Improvements(1)
Improved /mcp panel navigation so up and down arrow keys strictly navigate between MCP servers and plugins, left and right arrow keys cycle available actions, and Esc exits action mode or the panel.
Fixes(6)
Fixed duplicate agent entries appearing in the /agents picker when an agent with the same name is discovered across multiple configuration sources or plugins.
Fixed tool initialization and agent startup failing when launched from an inaccessible or deleted working directory by using absolute URI schemes for in-memory parameter schemas.
Fixed headless CLI invocations with piped standard output or standard error hanging on exit by setting FD_CLOEXEC on the preserved streams so child processes do not keep the caller's pipes open.
Fixed MCP configuration parsing failing when mcp_config.json contains single-line (//) comments, multi-line (//) comments, or trailing commas.
Fixed orphaned annotation files accumulating in local storage when deleting conversations.
Fixed /btw side questions in conversations with an active /goal being forced to continue and making unintended tool calls.
Model autocompletion improvements, subagent streaming optimizations, and MCP tool resolution fixes
Improves /model autocompletion with Tab acceptance, reduces subagent streaming telemetry overhead, and delivers stability fixes for prompt hooks, MCP tools in subagents, tool call IDs in request history, and OAuth token refreshes.
Improvements(2)
Improved /model autocompletion to accept the proposed model name ghost text using Tab.
Reduced subagent streaming overhead by sending subagent trajectory metadata once per subtrajectory instead of with every step.
Fixes(11)
Fixed commands with subcommands (such as models or agents) hanging on inherited standard input pipes.
Fixed CLI crashes caused by prompt hooks by catching hook panics and rejecting nil completion configurations in model requests.
Fixed tool invocations and results omitting tool-call IDs when reconstructing request history for Gemini models.
Fixed tool permission prompts for native and MCP tool calls displaying generic titles instead of human-readable action descriptions.
Fixed unconfigured or interrupted Google Cloud authentication dropping into broken chat sessions instead of reprompting the sign-in screen.
Fixed transient authentication errors from clock skew by proactively refreshing browser and WIF OAuth tokens 5 minutes before expiration.
Fixed instructional cycle mode placeholders reappearing in the input prompt after typing and clearing content.
Fixed subagents defined with enable_mcp_tools=true failing with unknown tool errors by ensuring the MCP dispatcher is available to the subagent executor.
Fixed prompts submitted immediately after login being discarded by queueing them until background account verification finishes.
Fixed cancelled or killed subagents remaining stuck in the "Running" state in the UI task drawer across ancestor conversations.
Fixed parsing MCP JSON configuration files on Windows when saved with a UTF-8 byte order mark (BOM).
Model selection command arguments, reasoning effort autocompletion, transient retry handling, and UI performance fixes
Adds direct /model switching with inline ghost text autocompletion, dynamic /effort hint completion, artifact filesystem event coalescing, and fixes for transient HTTP 502 retries, CPU redraw utilization, and Windows path resolution.
Improvements(3)
Added a /model argument to switch models by name, slug, or label and set defaults in one step with inline ghost text autocompletion.
Improved the /effort hint to complete typed text dynamically instead of displaying a static placeholder.
Improved artifact handling in sessions generating many files by coalescing bursts of filesystem events into a single rescan.
Fixes(8)
Fixed selectable reasoning effort configuration for Gemini 3.1 Pro and Gemini 3.5 Flash when authenticating via Gemini API keys.
Fixed continuous interface redraws when the tasks panel or subagent detail panel was open with no active tasks, significantly reducing idle CPU usage.
Fixed running subagent elapsed timers freezing on screen while the parent agent was in a waiting state.
Fixed transient HTTP 502 Bad Gateway errors terminating runs by adding automatic retry handling with backoff.
Fixed self subagents launched from unconfigured sessions re-resolving configurations from scratch and drifting from parent autonomous execution modes.
Fixed Windows file deletion sharing violations by retrying deletions with backoff before failing.
Fixed headless daemon startup banner printing unsupported localhost browser URLs to service manager logs.
Fixed the built-in migrate-workflows skill to properly handle Windows path separators and home directories.
Voice dictation, embedded ripgrep code search, status line cost metrics, and UTF-8 encoding fixes
Adds /voice dictation with mic-serve SSH forwarding, embedded ripgrep for consistent agent code search, custom status line cost tracking, granular script runner permission suggestions, and UTF-8 file editing and MCP error fixes.
Improvements(7)
Added /voice dictation to transcribe speech directly into the prompt with f5 and /voice toggle shortcuts, introduced the mic-serve subcommand to forward a local microphone over SSH, and improved quota messaging for concurrent stream limits.
Added a cost field to the status line data model, exposing unrounded estimated session costs to track running token spend.
Improved agent code search by bundling an embedded ripgrep binary instead of invoking external system binaries, increasing search performance and cross-platform consistency.
Improved the always-proceed permission mode to auto-approve MCP tool calls and page reads without prompting.
Improved allow-always permission suggestions for script runners (npm run, yarn, pnpm, cargo run) by scoping approvals to specific script names.
Improved conversation titles by generating them automatically on creation, providing meaningful session names in the resume picker.
Improved error diagnostics when an MCP server configured for Google credentials cannot locate Application Default Credentials (ADC).
Fixes(4)
Fixed corrupted file edits in documents containing non-ASCII text (CJK characters, accented letters, emoji) caused by inexact offset matches producing invalid UTF-8.
Fixed sessions stalling mid-response when a tool result or file diff contained invalid UTF-8.
Fixed file writes being reported as failures after successfully saving to disk, preventing unnecessary agent retry loops on large files and notebooks.
Fixed precedence resolution so explicitly configured skill and plugin paths take priority over automatically discovered customizations during name collisions.
Skill icon branding, empty directory indexing in @-completion, and headless exit code fixes
Adds skill icon branding with Unicode alignment in catalogs and slash commands, indexes empty directories in @ path completion, auto-approves workspace-scoped read access in review mode, and fixes headless print mode error handling.
Improvements(4)
Added skill icon and visual branding support across the CLI, displaying emoji icons declared under metadata.icon in SKILL.md frontmatter across the /skills catalog list view, detail inspection headers, and slash command autocompletion popups with multi-byte Unicode display width calculation.
Improved @ file path autocompletion by indexing empty directories alongside files in ripgrep search results, enabling directory-only and unpopulated structures to be traversed.
Improved permission management by automatically granting workspace-scoped read access under the default review mode, removing repetitive approval prompts for reading or listing workspace files while strictly maintaining confirmation for modifications and external access.
Improved Git repository inspection performance by skipping recursive submodule worktree scans while continuing to track commit pointer updates.
Fixes(5)
Fixed print mode (-p/--print, including --output-format json and stream-json) treating benign tool execution errors and permission denials as fatal run failures with non-zero exit codes.
Fixed the CLI overwriting and discarding unparsed configuration in settings.json when encountering an unrecognized setting value or syntax error on startup.
Fixed /skills, /plugins, /agents, and /hooks commands reporting that no customizations were found when invoked without an explicit agent configuration by mounting built-in customizations.
Fixed CJK draft jitter in the main prompt by normalizing boundary whitespace in streaming transcript updates.
Fixed the conversation spinner animation loop continuing to tick in the background after turn completion, eliminating idle CPU wakeups.
Dynamic remote control port allocation, banner logo toggles, and rendering flags
Introduces automatic free port selection for --remote-control, environment controls to suppress logo art for narrow terminals and screen readers, and renderer bypass options.
Improvements(2)
Added the AGY_CLI_HIDE_LOGO environment variable for narrow terminals, screen readers, and recordings to suppress logo art while retaining version and account details.
Added AGY_CLI_DISABLE_ESCAPE_SEQUENCE_OPTIMIZATIONS to bypass the renderer's dirty-rectangle and diffing optimizations.
Fixes(1)
Fixed --remote-control failing when a default port was occupied by automatically binding to an available port provided by the operating system.
Project name flags, customizable picker keybindings, typo-tolerant @-completion, and audio attachments
Adds support for passing project names to --project, customizable keybindings for conversation management, typo-tolerant @ file completion, and standard audio attachments alongside reliability and terminal layout fixes.
Improvements(6)
Added support for passing a project name to --project, which previously accepted only a project ID and failed on anything else.
Added item.rename and item.delete keybindings for the conversation picker's rename and delete actions, allowing F2 and F4 to be rebound in keybindings.json on keyboards without function keys.
Improved @ file path completion with a typo-tolerant fallback, so a query with a transposed or mistyped character still finds the file without outranking exact matches.
Improved audio attachments by recognizing standard formats Gemini models accept, including .wav, .mp3, .m4a, .aac, .flac, and .opus.
Improved keystroke responsiveness on Windows by discarding key-release events reported by the console, cutting rendering work per keystroke.
Improved the artifact viewer footer by placing the outline shortcut next to scroll and page hints for clearer document navigation.
Fixes(6)
Fixed print mode (-p) exiting successfully with an empty response when the agent state stream dropped mid-run; it now surfaces the stream error and exits non-zero.
Fixed a valueless prompt flag swallowing the next flag as its prompt (such as --print --sandbox "do the task"), turning both that and stray trailing arguments into explicit errors.
Fixed expanded /btw cards pushing footer hints and prompt off-screen on long answers by clamping output to terminal height with scroll support.
Fixed /resume opening below the fold instead of the active workspace when running the CLI from a subdirectory.
Fixed text losing styling after file links where the link's styling reset also cleared surrounding prose formatting.
Fixed stray characters appearing in the prompt every few seconds by restricting periodic input-mode re-arming to terminal multiplexers.
Consolidated agent execution harness and media attachment fixes
Consolidated agent execution harness for consistent tool, hook, and prompt behavior, alongside fixes for slash command visibility, Vim insert mode task navigation, and Ogg audio/video attachment MIME types.
Improvements(1)
Improved the agent execution harness by consolidating onto a single execution path, giving more consistent tool, hook, and prompt behavior.
Fixes(3)
Fixed /teamwork-preview and some other slash commands disappearing for some users.
Fixed Enter not opening an active background task or subagent while the prompt was in Vim insert mode.
Fixed attaching Ogg audio and video files such as .ogg, .opus and .ogv, which the model rejected because they were sent as the generic application/ogg.
mcp management subcommands (add, remove, list, enable, disable) for mcp_config.json, debounced @ file path completion powered by bundled ripgrep, Gemini API key reasoning effort adjustment for Gemini Flash models, and comprehensive fixes for binary MCP resource offloading, Workforce Identity Federation token refresh, /mcp configuration preservation, Kitty keyboard protocol handling, /btw with custom agents, and finished task status tracking.
Improvements(3)
Added mcp subcommands (add, remove, list, enable, disable) for managing MCP servers in your user-level mcp_config.json without hand-editing it, covering both stdio and HTTP servers through --type, --env and --header.
Improved @ file path completion in large workspaces by running the lookup through the bundled ripgrep and debouncing keystrokes, and by ranking a file whose name matches your query above a directory or generated artifact that merely contains it.
Improved /effort so it adjusts reasoning effort for Gemini 3.6 Flash and Gemini 3.7 Flash when you sign in with a Gemini API key, a route that previously reported those models as not adjustable even though the same models were adjustable on every other sign-in path.
Fixes(15)
Fixed links printing as raw escape sequences on terminals that do not implement OSC 8 hyperlinks, such as Terminal.app, and extended the same detection to command output and alert bodies, which still emitted hyperlinks unconditionally.
Fixed a = character accumulating in the prompt every couple of seconds on terminals that do not implement the Kitty keyboard protocol, where the renderer's periodic re-arming of that protocol was printed as literal text instead of being interpreted.
Fixed @ file path completion showing files that were in your .antigravityignore.
Fixed the artifact list showing a percent-escaped filename such as quarterly%20plan.md for artifacts whose name contains a space, so the list row, the inline preview and the detail header all spell the name the same way.
Fixed the prompt editor's cursor becoming misaligned when text wrapped.
Fixed the /mcp panel dropping enabledTools, timeoutSeconds, url and tools.eager from mcp_config.json when you toggled a server on or off; it now preserves any field it does not recognize, so configuration written by a newer client survives an edit.
Fixed read_resource discarding non-image binary content returned by an MCP server, which now offloads every blob to disk and inlines only small text and image resources, so large PDFs, audio and other binary resources are usable instead of silently dropped.
Fixed Workforce Identity Federation sign-in signing you out roughly every hour, because the token refresh went to the standard sign-in endpoint rather than the federated one.
Fixed skills that ship with the CLI being reported as not built-in, so the builtin flag in the /skills listing under --output-format json is accurate and the /skills panel groups them correctly.
Fixed the reasoning-effort description under the timeline gauge in the /effort and /model pickers overflowing on narrow terminals, where it is now omitted so the gauge stays readable.
Fixed /btw failing with a planner configuration error in conversations driven by a custom or SDK-defined agent, so side questions work regardless of which agent you are running.
Fixed a failed /btw side question always reporting model returned an empty response, which was a hardcoded guess rather than the real failure; the card now shows the side question's own error when there is one.
Fixed the status line, the active-items list and the /tasks panel continuing to show background tasks that had already finished.
Fixed the CLI overwriting a settings.json it could not parse with default settings, which silently reverted every setting the next time anything was saved; a refused save now leaves the file byte-identical so you can repair it by hand, and the status line names the file.
Fixed a subagent whose definition leaves model at inherit failing to start when the parent agent has no model of its own, which now falls back to the default fast tier.
Stream-JSON print input mode for persistent session drivers, direct rule file configuration in markdown agent frontmatter, plugin rule support via rules.json, current tool call hints on the spinner, and fixes for allocator preloading on Cloud TPU VMs, keyring credential restoration, terminal scrolling and unicode display in the model picker, and artifact subdirectories.
Improvements(4)
Added --input-format stream-json to print mode, which reads newline-delimited JSON prompts from stdin and runs one turn per message in a single conversation, so a driver can keep a session open.
Added a rules: key to markdown agent frontmatter, so an agent can name rule files directly instead of inheriting your whole rule tree; a rule named this way always applies.
Added plugin support for a top-level rules.json, so a plugin can declare the rule files it ships the same way it already declares its skills.
Added a more specific hint on the spinner line, which now names the tool call the agent is currently running.
Fixes(10)
Fixed the CLI aborting at startup on hosts that preload their own allocator through LD_PRELOAD, such as Cloud TPU VM images, where it died before the interface appeared.
Fixed personal accounts hitting a resource-exhausted error at startup when saved credentials were restored from the system keyring, and cut the repeated account checks that followed.
Fixed your billing project, and the license tier it selects, being lost when the CLI restarted and restored saved credentials from the system keyring.
Fixed a spurious "Out of credits" message for enterprise users signed in through Application Default Credentials.
Fixed the /model picker cutting off models, and its own header, on terminals too short to show every model; the list now scrolls.
Fixed the /model picker misaligning its columns for model names that are not plain ASCII, by measuring column widths in display cells rather than bytes.
Fixed the spinner line flickering to a generic label between quick tool calls.
Fixed streamed text corrupting non-ASCII characters into replacement characters, in both the interactive display and --output-format stream-json text deltas.
Fixed artifacts written into a subdirectory of a conversation's artifact directory, such as scratch/, never appearing in the artifact list.
Fixed reading a .wav file failing as an unsupported media type, by normalizing non-canonical MIME types such as audio/wave before they reach the model.
Enterprise sign-in fast path, MCP OAuth metadata documents, /context scrolling, and markdown agent customizations
Fast return path for enterprise sign-in, support for OAuth client ID metadata documents in MCP servers, terminal-sized scrollable /context and /usage panels, unified inheritCustomizations configuration for markdown agents, read-only permissions outside the workspace, and fixes for language server crashes, hyperlink rendering, artifact approval tracking, MCP error isolation, and bracketed paste.
Improvements(5)
Added a faster return path for enterprise sign-in, offering returning users a single Continue with ... option naming the authentication method they last used.
Added support for OAuth client ID metadata documents when connecting to an MCP server, allowing servers implementing this specification to connect without requiring a manually supplied client ID or dynamic client registration.
Improved the /context panel, making it scrollable and dynamically sized to the terminal, with /usage sized the same way.
Improved markdown-defined agents with a single inheritCustomizations switch that controls whether an agent adopts skills, rules, plugins, subagents, and MCP servers, replacing inconsistent per-kind defaults.
Improved workspace boundary settings: accessing paths outside the workspace now grants read access only, with writes requiring authorization based on the active execution mode.
Fixes(9)
Fixed the CLI exiting silently when its language server failed at startup or during execution; failures are now logged to the terminal and exit with a non-zero status.
Fixed sign-in URLs and hyperlinks rendering as garbled escape sequences on terminals that do not support OSC 8 hyperlinks (such as GNU screen), detecting unsupported terminals and rendering plain text instead.
Fixed Enter keystrokes being swallowed while the path-suggestion popup was loading, which previously sent the prompt and popped results open above it.
Fixed the artifact list marking previously unreviewed, commented, or rejected artifacts as approved when reopening the list.
Fixed malformed MCP server configurations causing all other servers to fail; invalid entries are now logged and skipped while remaining servers continue loading.
Fixed built-in skills, rules, and plugins being stripped when an agent configured inherit_user: false, ensuring built-in capabilities remain available while personal customizations are excluded.
Fixed conversation branching and rollbacks failing with too many SQL variables on long trajectories.
Fixed conversation titles unexpectedly jumping to older messages after trajectory truncation by protecting the title checkpoint.
Fixed multi-line pastes submitting individual prompts per line on terminals that do not support bracketed paste.
Direct GEMINI_API_KEY authentication support, /codesearch fallback resilience, cached artifact list scrolling, manage_task access for custom declarative agents, embedded ripgrep caching with SHA-256 verification, and critical fixes for trajectory truncation, database growth, transcript race conditions, and subagent path validation.
Improvements(10)
Added support for GEMINI_API_KEY, allowing the CLI to run directly against the Gemini API without signing in. Set modelProvider: "gemini" in settings.json, export GEMINI_API_KEY, and optionally configure GOOGLE_GEMINI_BASE_URL for custom endpoints. The banner and /help show Gemini API key as the credential, and /logout explains that it originates from the environment.
Improved /codesearch resilience by falling back to a local search when the ripgrep binary cannot be executed, such as when blocked by endpoint security software.
Improved artifact list scrolling performance by caching each file's preview line count instead of re-reading every listed file from disk on every keypress.
Improved custom agent support by making manage_task available to declarative agents that have not opted out of fundamental components, enabling custom agents to list and kill background tasks, with clearer step descriptions like Killed task X and Checked task X.
Improved search_web step visualization by displaying the query while the search is in progress rather than only after completion.
Improved agent behavior after backgrounding tasks by removing anti-polling reminder text from manage_task and manage_inbox tool responses to prevent unintended polling loops.
Improved the /resume picker by clearly relabeling its second tab.
Improved deleting conversations in the /resume picker by remapping the shortcut from ctrl+delete to f4 for reliable cross-platform terminal compatibility on macOS.
Improved the load time when opening large conversations by fetching step headers in batches without reading each step's full payload.
Improved embedded ripgrep reliability and security by extracting binaries into the user cache directory instead of /tmp, adding SHA-256 integrity verification, and using atomic renaming to prevent concurrent execution races.
Fixes(11)
Fixed distorted startup banner rendering in Cloud Shell by automatically enabling a conservative rendering mode for web terminals.
Fixed direct Gemini API mode getting stuck on the sign-in screen instead of launching into the main UI, and resolved failures with custom GOOGLE_GEMINI_BASE_URL endpoints by omitting unsupported session fields.
Fixed conversations disappearing from the /resume picker after switching tabs, and preserved the tab hint in the shortcut bar even when a tab is empty.
Fixed trajectory truncation over-pruning conversation history by charging the size budget solely against reclaimable steps rather than protected checkpoints or already-cleared residues.
Fixed unbounded growth of the SQLite conversation database during background task and subagent wakeups caused by duplicate grants and settings being appended to persisted rows.
Fixed transcript corruption caused by background messages appending while context compaction was rewriting the log.
Fixed an issue where a hung sandbox initialization prevented subsequent messages from being processed.
Fixed path traversal vulnerability in define_subagent by strictly validating model-supplied agent names before creating directories.
Fixed artifact list rendering bugs involving wide CJK characters and emoji, boundary handling when entries disappear, and viewport scrolling overflow.
Fixed Esc key in the artifact viewer triggering duplicate review submission confirmations for previously completed reviews.
Fixed missing tool call line rendering in collapsed/expanded tool groups for invoke_subagent, send_message, manage_subagents, define_subagent, and generate_image.
Artifact outline viewer navigation, print-mode read-only slash commands, machine-readable output for models and agents, alert callout and carousel rendering in markdown artifacts, terminal hyperlink improvements, and fixes for headless --mode, startup diagnostics, subagent messaging, and quota reporting.
Improvements(13)
Added a heading outline to the artifact viewer, opened with t, so a long markdown document's structure is visible at a glance and you can jump straight to a section instead of scrolling.
Added non-interactive answers for more read-only slash commands in print mode, so -p "/permissions", /hooks, /help, /changelog and /config each emit one tab-separated record per line — or a structured payload under --output-format json and stream-json — without starting an agent turn, spending quota or leaving a conversation behind, with /help listing exactly the commands print mode answers and /changelog printing the release notes.
Added machine-readable output to the models and agents subcommands through an --output-format flag accepting json and stream-json, and moved their error messages and progress spinner off stdout so captured output contains only the list.
Added a disable-slash-command: true flag for a skill's SKILL.md frontmatter, which hides that skill from the / menu and from /name resolution while leaving it discoverable and invocable by the model, so a large skill library no longer floods the command menu.
Added rendering for alert callouts ([!NOTE], [!TIP], [!IMPORTANT], [!WARNING], [!CAUTION]) and for carousel blocks in markdown artifacts, which previously showed as raw markup.
Improved terminal hyperlink support, enabling clickable links over SSH and in hyperlink-capable tmux, keeping word-wrapped URLs clickable past their first line, linking URLs that the markdown renderer had left as plain text including ones inside code blocks, no longer highlighting unrelated links together on hover, and fixing horizontal scrolling and text selection on lines that contain a link.
Improved tool call headers so every native and MCP tool call carries a short summary of what it is doing instead of a bare name or a raw argument blob, with the same summary naming background tasks and subagents in the activity list and the matching action phrase describing the request in permission prompts.
Improved headroom for large toolsets by raising the per-session limit on tool declarations, so heavy MCP, plugin and skill setups stop being rejected for having too many tools.
Improved sign-in with Application Default Credentials by resolving the quota project automatically, so service-account logins work without extra configuration.
Improved repository detection so a Git repository nested inside a larger multi-repository checkout resolves to the intended root, with submodules and worktrees covered.
Improved headless -p runs so the agent settles a choice itself where it would otherwise ask, instead of stalling on a question nobody is there to answer.
Improved the schedule tool by dropping the redundant Timer Cancelled notification when a one-shot timer with an early-termination condition is cancelled, and by naming both the condition and the exact sender whose message satisfied it in the step result.
Improved file paths in tool step lines by collapsing your home directory to ~ instead of truncating the front of the path, so the interesting end of a long path stays on screen.
Fixes(13)
Fixed --mode being ignored in headless -p runs, where a valid value such as accept-edits or plan was never applied and an unrecognized value produced no warning at all.
Fixed startup diagnostics being swallowed into the log file instead of reaching the terminal, including crash notices, the --conversation not-found warning, the --print and --prompt-interactive conflict, conversation load errors, and the --mode and --agent warnings.
Fixed a character disappearing from your submitted prompt where it wrapped at the terminal's right edge, caused by the echoed prompt being indented after it had already been wrapped to the full width.
Fixed file citations in a response losing their line numbers, so a cited range renders as path:10-25 again instead of just the file name.
Fixed a write-in answer in a multi-select question discarding the boxes you had already ticked, and fixed a stale write-in being submitted after you went back to a predefined option.
Fixed large background tasks crowding out the prompt by limiting each active item below the input box to a single line.
Fixed corruption of config.json by writing user config atomically, so a crash or a concurrent writer can no longer leave a truncated file that silently breaks settings persistence.
Fixed the CLI giving up on a slow OS keyring after one second and falling back to empty storage, which forced a re-login; it now waits five seconds, as every other keyring operation already did.
Fixed conversation preview titles derived from the first user message by cutting them at the first line and capping them at 500 characters, so a large pasted or scripted prompt no longer produces a multi-megabyte stored title.
Fixed a crash on Windows when resolving the conversation transcript path, by honoring the path's drive letter in the trajectory log artifact converter.
Fixed a subagent going silent on its parent when its own message failed to send, and fixed the parent being handed an empty notice when a subagent went idle without a final response.
Fixed the read_url_content tool leaking a connection on every call, which could eventually exhaust the machine's available ports.
Fixed the status line reporting model quota that was always one fetch out of date, so it now matches the quota you actually have left.
Vim modal editing mode, Vim-aware submission, print-mode slash-command enhancements, and various CLI and permission fixes
Vim modal editing mode with Normal, Insert, Visual, and Visual Line modes, Vim-aware prompt submission, Vim editing in comment/diff views, print-mode slash-command and non-interactive enhancements, plugin configuration improvements, and fixes for command auto-approval, MCP servers, and autocomplete history.
Improvements(9)
Added a Vim editing mode, off by default and switched on from /settings under Editor Mode, bringing modal editing to the prompt with Normal, Insert, Visual and Visual Line modes, a mode badge in the status line and a Vim tab in /help, and covering the h/j/k/l, w/b/e and W/B/E motions, the 0/^/$/gg/G boundaries, the f/F/t/T character searches with their ; and , repeats, the i/a/I/A insert entries, the d/c/y operators alongside x, D and C filling the unnamed register for p, and the iw/aw/iW/aW/ip/ap text objects.
Added Vim-aware submission so a prompt can be sent without leaving modal editing, through an Editor Mode > Insert First toggle that opens the prompt in Insert mode where a bare Enter submits and a modified key such as shift+enter or ctrl+j inserts a newline, ctrl+s and ctrl+enter keys that submit from Normal mode, an Enter binding available as vim.insert.submit, and the full set of vim.* scopes in keybindings.json for remapping any of it.
Added Vim editing to the comment editors in the diff and artifact-detail views, whose footer hints follow the active mode rather than showing one fixed set of bindings, and made a collapsed block behave as a single unit under Vim motions.
Added non-interactive answers for the read-only slash commands in print mode, so -p "/usage", /quota, /credits, /model, /effort and /skills emit one tab-separated record per line — or a structured payload under --output-format json and stream-json — without starting an agent turn, spending quota, or leaving a conversation behind.
Added an explicit refusal for the remaining interactive-only slash commands in print mode, which previously fell through as literal prompt text and let the model answer as though the command had run, so -p "/clear" reported the context cleared while nothing was cleared; each now fails with the flag or subcommand that replaces it.
Improved plugin enable and disable so config.json is the only place enablement lives, seeded once from each plugin's manifest, which stops a plugin that later ships "disabled": true from switching itself off under someone who was already running it and stops a shipped-default change from moving every user on the next release.
Improved the artifact detail view by wrapping long lines in code files instead of clipping them at the right edge.
Improved model retries by honoring the server-supplied retry delay instead of the client's own backoff, so a retry after a rate limit or an overload waits exactly as long as the server asks.
Improved model-loading errors so a permission failure such as a missing license or a missing IAM role is shown as itself, with a troubleshooting link, instead of a generic failure during the run.
Fixes(9)
Fixed an allowlist entry that tokenizes to zero command words — command(time), a comment-only entry, or an empty compound such as () — matching every command and silently auto-approving anything the agent ran; such an entry now matches nothing.
Fixed commands being auto-approved while the session was in request-review or strict permission mode.
Fixed admin controls being skipped for MCP servers at startup, where a fetch made before authentication cached "admin controls not applicable" and allowed every server for the next five minutes, and fixed the built-in Chrome DevTools MCP server being blocked outright by admin controls.
Fixed MCP progress callbacks being dropped and the task log file not being initialized, so long-running MCP tool calls report progress again.
Fixed a slash command receiving the collapsed [Pasted text #N] placeholder instead of the pasted content when its argument was pasted into the prompt.
Fixed the prompt saving the typed prefix rather than the completed command name to history when an autocompleted slash command was submitted, so up-arrow recall replays what actually ran, and made an alias require an exact match before auto-executing so a partial alias fills the prompt instead of running a command you did not finish typing.
Fixed an asynchronous settings refresh re-enabling the feedback survey partway through a print-mode run that had switched it off.
Fixed spurious "Out of credits" errors, where an empty credits response was read as a balance of zero.
Fixed the "Use AI Credits" setting being offered to accounts signed in through a Google Cloud project or application default credentials, where it does not apply.
Gemini Enterprise & WIF authentication, read-only .git sandbox rules, hook ordering improvements, and subagent and CLI fixes
Gemini Enterprise Business sign-in, Workforce Identity Federation (WIF) and Application Default Credentials (ADC) auth support, read-only .git sandbox access, hook ordering improvements, and fixes for subagent tree termination, --model/--effort flag resolution, and multi-select spacebar toggling.
Improvements(7)
Added Business sign-in for Gemini Enterprise accounts, so you can authenticate with a Google Cloud project under Google Cloud terms, use a license seat allocated or auto-assigned from your organization's GE-Standard or GE-Plus subscription, run inference in a chosen region, and have your organization's administrator controls applied to various features.
Added Workforce Identity Federation sign-in for enterprise users, available as the Use advanced SSO config option on the Google Cloud sign-in screen, so organizations that federate identity through an external provider can authenticate with their own identity provider.
Added sign-in with Application Default Credentials so you can use Agent Platform.
Added a non-blocking advisory banner when the same conversation is already open in another CLI instance on the same machine, pointing at /fork so two sessions no longer interleave writes into one trajectory.
Improved the terminal sandbox by granting read-only rather than writable access to a Git repository's .git directory, so the agent can inspect repository metadata without being able to rewrite it from inside the sandbox.
Improved hook ordering so hooks defined in hooks.json run before the built-in termination checks, which lets PostInvocation hooks observe the final invocation of a turn and lets Stop hooks run at all instead of sitting unreachable behind the built-ins.
Improved the schedule tool to accept DurationSeconds and MaxIterations when a model emits them as bare JSON numbers rather than strings, accepting integral values and rejecting non-integral ones with a clear error instead of failing the call.
Fixes(11)
Fixed --model and --effort being ignored in interactive sessions and in headless -p runs, where the flags were applied after model configuration had already been initialized so the run silently fell back to the persisted or default model.
Fixed a bare --effort resolving against the default model instead of the model you actually have selected, which could silently move you to a different model.
Fixed stopping a subagent tree stopping only the conversation it was invoked from, while every descendant subagent and the background tasks they owned kept running and the CLI still reported them as killed.
Fixed a forced-continuation deadlock where a coordinator waiting on active subagents or background tasks would loop injecting empty continue steps until it hit the invocation limit, wasting tokens and blocking progress.
Fixed the spacebar not toggling an option in multi-select prompts, including the ask_question dialog and the onboarding import checkbox, which left x as the only working toggle; the hint bar now advertises it.
Fixed the Left and Right arrow keys being captured to navigate the input box suggestion dropdown, so you can move the cursor and edit text again while suggestions are showing.
Fixed the model picker's "No models available" state rendering without its header and footer, so it now shows the standard chrome and an esc hint to go back.
Fixed tools that an MCP server marks to always run in the background executing as blocking calls that stalled the turn.
Fixed an MCP process leak when a server connection dropped unexpectedly.
Fixed the artifact viewer corrupting plain documents by horizontally clipping every document rather than only the diagram artifacts that need it.
Fixed the sandbox not recording blocked network requests when the command itself exited successfully, which hid the fact that a request had been denied.
Slash-command expansion in print mode, non-blocking MCP startup, session-wide permission memory, system temp directory write grants, and hook and UI stability fixes
Added slash-command and skill expansion in print mode, non-blocking background MCP loading during interactive startup, session-scoped permission pattern recording, system temporary directory write grants, and stability fixes for stop/PostToolUse hooks, artifact viewer, and MCP authentication.
Improvements(4)
Added slash-command and skill expansion to print mode, so a headless run such as -p "/my-skill review this diff" now resolves and applies the skill instead of sending it as literal text, with --disable-slash-commands to opt out.
Improved interactive startup so a slow or hanging MCP server no longer stalls the first agent turn, loading MCP servers in the background for the interactive session while headless and one-shot runs keep blocking so their single scripted turn still sees the full toolset.
Improved permission grants so a pattern approved at a prompt is recorded for the rest of the conversation, letting later commands that match it run without prompting again.
Improved the default system temporary-directory grant to cover writes as well as reads, so agents no longer trigger a permission prompt when creating or updating files there.
Fixes(7)
Fixed stop hooks that always block hanging the agent forever; after a configurable number of consecutive continuations, the hook can no longer block and the turn ends normally.
Fixed PostToolUse hooks firing on non-tool steps such as user input and model responses, which also caused them to ignore their configured matchers.
Fixed slash commands not being recognized when separated from their arguments by a newline or tab, so a prompt starting with a command followed by a newline is now parsed as a command instead of being sent verbatim.
Fixed deleting into a collapsed paste placeholder removing one character at a time, which left a visible fragment in the prompt while the full pasted content was still submitted; the block is now deleted atomically.
Fixed the artifact viewer losing syntax highlighting when returning from the editor view, and returning to the wrong panel when exiting the artifact detail view.
Fixed the headless stream-jsoninit event advertising tools that are not available in your build.
Fixed MCP servers forcing a full re-authentication after a dropped connection.
Added structured output formats (json, stream-json) for print mode, support for custom JSON schema validation, enriched tool and subagent payloads, copyOnSelect configuration setting, and improved compound-command permission rules.
Improvements(7)
Print mode (-p / --print) now supports structured, machine-readable output via the --output-format flag (text (default), json, or stream-json), so headless runs in CI, eval harnesses, and scripts can consume the CLI's output programmatically; these flags are now discoverable in --help.
Added the stream-json output format: a strongly-typed NDJSON event stream that emits typed init, step_update, and terminal result events with a stable, closed-vocabulary step_type discriminator, so consumers receive progress incrementally instead of waiting for the whole run to finish.
Added the --json-schema flag to enforce a custom JSON schema on the structured output, accepting either an inline schema string or a path to a schema file; for stream-json the schema applies to the final result event.
Enriched the structured stream with a tool_info object for each tool call (canonical tool name, parameters, and output) and a subagent_info payload for delegated subagents (including conversation_id and log_uri) so consumers can correlate child trajectories.
The JSON usage object emitted by json and stream-json now reports token accounting including cache_read_tokens, so non-interactive consumers can attribute prompt-cache hits.
Added a copyOnSelect setting (default on, toggleable in /settings) that controls whether releasing a mouse text-selection auto-copies it to the system clipboard in the TUI's altscreen rendering mode; disable it to stop the automatic copy on release — useful when the auto-copy is unwanted or corrupts certain payloads.
Improved compound-command permissions so an exact chained command (such as git fetch && git rebase) can be saved as an allow-always rule and no longer re-prompts on the next identical run.
Permission prompts for compound commands, and fixes for disabled plugins, MCP OAuth, and clipboard corruption
Improved permission prompts for compound shell commands, and fixed disabled plugins running hooks, MCP OAuth issuer validation, CJK clipboard copying on Windows, and /btw error on startup.
Improvements(1)
Improved permission prompts for compound shell commands so the full command is shown when any part of it needs approval.
Fixes(5)
Fixed disabled plugins still running their hooks and contributing other customizations, which could keep a broken hook active and break file-editing tools even after the plugin was turned off.
Fixed MCP OAuth against providers that do not strictly follow the spec (such as Salesforce and Atlassian) by relaxing issuer validation and including the refresh_token grant.
Fixed /btw failing with a "parent conversation not found" error when used as the very first action in a fresh session.
Fixed clipboard corruption of CJK and other non-ASCII text when copying on Windows.
Fixed print mode (-p) sending a prompt before the account-eligibility check finished.
Markdown format custom agents, /copy index support, progressive /codesearch streaming, default temp directory access, and stability fixes
Added support for defining custom agents in Markdown format (agent.md), optional index argument for /copy, progressive streaming in /codesearch, default read access to system temporary directory, and resolved various TUI jitter, artifact viewer, and sandbox fixes.
Improvements(7)
Added support for defining custom agents using Markdown files (agent.md) with YAML frontmatter and H1-delimited system prompts. Markdown agents support mainAgent, subagent, hidden, inheritMcp, and commandExecutionPolicy frontmatter fields for fine-grained control over agent behavior. Dynamically defined subagents (via define_subagent) now also write Markdown format so they resolve correctly on external builds.
Added an optional index argument to /copy so /copy copies the n-th most recent response to the clipboard, while /copy and /copy 1 still copy the latest.
Improved /codesearch to render results progressively as they stream in, showing a live count while loading and letting you cancel an in-flight search with Esc instead of blocking until the whole search finishes.
Improved default file access by granting read access to the system temporary directory out of the box, resolved correctly per platform, so agents no longer trigger permission prompts when reading temporary files.
Improved support for markdown-based custom agents so custom agent management and selection behave more consistently.
Improved customization discovery by sorting rules and discovered paths deterministically, preventing unstable prompt ordering and needless prompt-cache misses.
Improved overall reliability and stability across the CLI with additional hardening and fixes for intermittent failures in background tasks, print mode, and interactive flows.
Fixes(11)
Fixed switching from a custom agent back to the default agent via /agents, which previously failed silently and left the conversation stuck on the custom agent's persona.
Fixed a crash when a command was blocked by sandbox permissions before its output was captured, and cleaned up the permission approval and denial messages.
Fixed the artifact viewer emitting garbage escape bytes when cycling to image mode on terminals that are detected but cannot actually render Kitty graphics, such as iTerm2.
Fixed the first keystroke (such as Esc) being dropped when opening the first artifact view on some non-Kitty terminals.
Fixed conversation jitter and a stranded input box during streaming so transient markdown reflow no longer shifts the pending line and input box upward.
Fixed print mode (-p) surfacing the real conversation-creation failure instead of a misleading "no active conversation" error.
Fixed the message list dropping its header when rewinding or resetting conversation steps.
Fixed a background auto-updater double-spawn race where two processes could each spawn an updater within a single update window.
Fixed sandbox error reporting so blocked actions are recorded even when the network proxy is disabled.
Fixed the screen going blank after the authentication page.
Fixed the ctrl+b shortcut being hardcoded to background shell commands even when none were running, so a remapped ctrl+b is now respected whenever there are no running shell commands in the conversation.
Reasoning effort command, model slug pinning, subagent model configuration, and TUI stability improvements
Added /effort command and --effort flag for reasoning effort selection, stable model slugs, subagent model config, redesigned /model picker, scrollable /settings panel, and various MCP and background task fixes.
Improvements(8)
Added a /effort command to view and change the current model's reasoning effort, with a left/right timeline-gauge picker and a direct /effort <level> form so you can trade latency for depth on the fly.
Added an --effort flag to select a model's reasoning-effort variant when launching the CLI.
Added stable, user-facing model slugs that appear in the /model picker and are accepted by --model, so you can pin a specific model reliably across sessions.
Added a model option to custom agent frontmatter so an agent runs at a chosen model tier (such as flash or pro) when invoked as a subagent, defaulting to inherit (the parent's model).
Redesigned the /model picker to group models by their base model and choose reasoning effort from a timeline gauge navigable with Left and Right, and added an effort badge to the status line for models that expose multiple effort variants.
Improved the /settings (/config) panel by making it a bounded, scrollable list so it renders correctly in short terminals instead of overflowing, and stopped it from flickering when opening and closing dropdowns.
Improved background-task reliability by moving long-running work onto a shared lifecycle with deterministic startup and shutdown and panic-safe launching, so a failure in one background task no longer disrupts the session and pending analytics are flushed on exit instead of dropped.
Improved responsiveness of bursty background refreshes by coalescing rapid repeated triggers into a single run, cutting redundant work.
Fixes(5)
Fixed a crash when triggering Authenticate on a remote MCP server in the /mcp panel.
Fixed MCP tool results containing embedded resources being silently dropped, so text and inline media returned by MCP servers now surface in the conversation.
Fixed permission checks splitting a single command into a pipeline when an argument contained quoted shell metacharacters (such as --grep="a|b"), which caused spurious permission prompts.
Fixed the file-view and file-search tools failing with invalid-UTF-8 errors when a multi-byte character was split at a truncation boundary.
Fixed a data race when collecting customization rules by guarding the shared structures.
Slash command chaining, diff scrolling fixes, and headless policy enforcement
Added slash command stacking (chaining multiple commands in a single prompt), diff viewer scroll fixes, headless mode settings.json policy enforcement, and subagent declaration fixes.
Improvements(2)
Added support for stacking multiple leading slash commands in a single prompt, so a chain like /plan /grill-me <prompt> parses, activates, and renders every command in the order you typed them.
Improved scrolling in the /diff viewer so paging through a diff no longer jitters or pushes the status line off the screen when lines wrap or comments expand.
Fixes(5)
Fixed custom agents that declare subagent: false still appearing in the available-subagents list and being invocable as subagents.
Fixed headless (-p / --print) runs so they now honor persisted settings.json policies, including permissions, file access, sandbox mode, auto-execution, and artifact review.
Fixed /btw side-questions leaking into the conversation list as duplicate entries that carried the parent conversation's title.
Fixed the prompt to honor a custom Enter binding to prompt.insert_newline, so a remapped Enter inserts a newline instead of submitting.
Fixed eligibility error messages so the CLI shows the real reason again instead of defaulting to a generic "unknown reason".
Codesearch command, copy-on-select, and performance optimizations
New /codesearch command for regex workspace searches, copy-on-select, async skill discovery, and stability fixes.
Improvements(7)
Added a /codesearch command (aliases /cs and /search) to interactively search code across your workspace, interpreting queries as regex by default with -F/--literal for exact matching and f:/file: globs to include or exclude paths.
Added copy-on-select in no-flickering mode so dragging highlights text and releasing the mouse copies the ANSI-stripped selection to the clipboard, and hides the virtual scrollbar so it no longer interferes with copying multi-line output.
Added an indicator at each context-compaction boundary so you can see where earlier compaction happened.
Improved interactive startup latency by loading skills asynchronously so the CLI no longer blocks on a synchronous, filesystem-heavy skill-discovery pass during bring-up.
Improved eligibility error handling by showing errors with a verification URL inline in the input loop instead of stacking them above the screen.
Improved customization loading latency for skills, rules, agents, and hooks by consolidating directory walks and caching filesystem lookups to cut redundant I/O during discovery.
Removed the padding spaces around inline code for tighter rendering.
Fixes(9)
Fixed code-block corruption where $..$ math expansion desynced from the Markdown parser and mangled fenced shell snippets such as git fetch "$GIT_REMOTE" by detecting fenced code blocks line-by-line.
Fixed headless (-p) runs hanging or silently auto-approving tools that require a permission confirmation, so the CLI now soft-denies such tools and prints a stderr notice naming the allow-rule needed to permit them.
Fixed outside-of-workspace file writes being incorrectly auto-approved in always-proceed mode.
Fixed high CPU and unbounded render cost on large conversations in no-flickering mode by making index rebuilds idempotent so the conversation index converges instead of growing on every rebuild.
Fixed lingering artifact comments after dismissing the artifact detail view and corrected no-flickering-mode row math so the status line renders correctly within the viewport.
Fixed repeated sign-in prompts on Linux caused by the OS keyring: the CLI now bypasses the keyring when no D-Bus session bus is present (headless hosts and containers), skips it for an hour after a timeout, and uses longer keyring timeouts so a slow-but-successful credential read is no longer cut short and forced into a fresh login.
Fixed MCP servers hanging the agent indefinitely when a server never responds by bounding connection, tool-listing, and per-tool-call attempts with timeouts.
Fixed conversations breaking after certain tool calls, which previously corrupted the conversation history and blocked all further responses.
Fixed customization rules being loaded twice when a rules directory is reachable through a symlink.
Full diff preview, OAuth print mode improvements, and lag fixes
Added full-screen diffs for file creations, OAuth authorization code pasting in print mode, and 5000+ step latency optimizations.
Improvements(3)
Added an f (full diff) shortcut to the create-file tool review screen so new-file confirmations can open a full-screen diff view, matching the existing file-edit experience.
Added support for pasting the OAuth authorization code in print mode (-p) via the controlling terminal (/dev/tty on POSIX and CONIN$ on Windows) when stdin is consumed by a piped prompt, and made truly headless runs fail fast with an actionable message instead of blocking.
Improved responsiveness on large conversations (5000+ steps) in no flickering mode by switching hot-path line-count methods to pointer receivers, cutting the per-frame prefix-sum cost and eliminating sustained 99% CPU and keystroke lag.
Fixes(6)
Fixed print mode silently downgrading to the default model when --model cannot be resolved by hard-failing with a non-zero exit and listing the available models, while interactive sessions keep the fallback-with-warning behavior.
Fixed permission checks not respecting the allowlist for nested command substitutions, so a command like echo "$(dirname $(git rev-parse --show-toplevel))" now runs without prompting when echo and git are allowlisted, instead of double-counting the nested command and prompting for review.
Fixed the CLI keybindings file staying out of sync with /keybindings when new default bindings are introduced by persisting the injected defaults while preserving user overrides.
Fixed garbled builtin tool headers such as CodeSearch(4 files found...) by mapping generic tool steps back to clean summaries like Read(/path) and CodeSearch(query).
Fixed mcp manager failing to resolve tool schema paths in standalone mode and leaking MCP server subprocesses after shutdown, which previously caused panics and cleanup failures for custom agents loading MCP tools.
Fixed a data race and copy-on-write violation when updating subagent states by cloning their stats before in-place mutation, preventing corrupted step counts and status for parallel subagents.
Custom agents flag, in-file keyword search, and Jujutsu diff fixes
Added --agent flag, in-file keyword search for artifacts, recursive subagent updates, and Jujutsu workspace diff fixes.
Improvements(6)
Added the --agent flag and agent/agents subcommand, allowing users to select a custom agent at launch and list available agents.
Added in-file keyword search (/) and jump navigation (n/N) to the artifact detail viewer, allowing users to find and cycle through matches without disrupting terminal escape sequences or image grids.
Added support for displaying nested subagents (grandchild and deeper) and handling tool confirmation requests across all subagent depths by recursively relaying nested subtrajectory updates to the root conversation.
Changed the default mode to respect write_file permissions allowlisted in settings.json under permission.allow, so pre-approved file writes no longer prompt for review.
Changed the default name for the newly initialized project to CLI Project for clearer workspace identification.
Improved the session exit output by placing the resume command on its own line, making it easier to copy and paste in terminals and tools like tmux.
Fixes(6)
Fixed print mode (--print / -p) silently exiting with a success code and empty output when a request failed server-side, now writing the error to stderr and returning a non-zero exit code.
Fixed agy -p hanging when run inside a shell script or subprocess by no longer reading stdin when a prompt is provided via a flag.
Fixed a data race on the /btw cancellation function.
Fixed interactive /diff viewer defects in Jujutsu (jj) workspaces by correctly prioritizing .jj over .git in colocated repos, fixing commit hash regex boundaries, and correctly highlighting active @ graph nodes.
Fixed workspace-local hooks defined in /.agents/hooks.json not loading after trusting a folder by reloading hooks whenever workspaces change.
Fixed misaligned markdown tables containing file links in chat output.
Request-review mode, execution mode cycling, and line-level diff previews
Public release of mode cycling, default request-review mode with interactive line-level diff previews, and settings panel integration.
Improvements(8)
Agent execution mode cycling is now publicly available: default -> accept-edits -> plan)
Added request-review (default) mode as the default execution behavior: automatically pauses before file write operations to display an interactive, line-level diff preview (f shortcut) where users can review, accept, or reject individual code modifications before they are saved to disk.
Added an Agent Mode option to the /settings panel so users can set and persist a default execution mode (default, accept-edits, plan) without manually editing settings.json or passing --mode on startup, with real-time synchronization so changes take effect immediately.
Added a dedicated "Create file" confirmation preview for new file creations (write_to_file without overwrite): renders new content as an addition-only diff preview.
Added /plan mode to replace legacy /planning, and removed /fast slash commands: consolidated and simplified execution mode switching around shift+tab mode cycling and the /plan mode prefix
Improved file-edit diff preview rendering: computed accurate line-level diffs with context lines (3 lines) and hunk separators, capped inline preview height with truncation hints, and added a comment confirmation prompt when exiting the diff view with unsent comments.
Improved UI footer keybinding hints across all panels (such as /tasks, /agents, /permissions, and /mcp) by replacing hardcoded hint strings with centralized layout helpers that dynamically respect customized global and local keybinding configurations (keybindings.json).
Improved the multiline conversation rename view in the /resume picker by dynamically adjusting input box width and padding, and right-aligning metadata columns (workspace, steps, time) on the top line to prevent horizontal scrolling or layout shifts during active renaming.
Fixes(6)
Fixed the tool confirmation dialog to accurately check normalized file URIs against active workspace directories, resolving an issue where valid in-workspace file creations and reads were incorrectly flagged with an "Reason: outside workspace" warning.
Fixed workspace initialization failures when launching the CLI inside dot-prefixed directories (such as .parent_dir/project ) by scoping path exclusion filters strictly to relative paths inside the workspace rather than rejecting dot-prefixed ancestor directories.
Fixed the /agents view header displaying agent.json instead of agent.md when creating new subagents.
Fixed the /agents panel's "Create New Agents" section displaying the wrong global configuration directory (~/.gemini/antigravity-cli/ instead of ~/.gemini/config/), ensuring users create global subagents in the location actively scanned during startup discovery.
Fixed statusline shortcut hints (? for shortcuts) and redundant escape hints (Esc to cancel) erroneously appearing inside full-screen overlay panels (such as /changelog, /artifact, and /settings) by correctly tracking overlay panel states.
Fixed inconsistent timestamp formatting in the /tasks panel and task detail views by converting agent-initiated background task timestamps (time.Time) from UTC to the local timezone.
Improved the /tasks detail panel to automatically scroll to the bottom as new background task logs stream in, and default to the latest output when opened while preserving scroll position if scrolled up manually.
Improved model generation resilience by adding automatic client-side retries when encountering transient errors.
Fixes(4)
Fixed dynamically defined subagents by transitioning definitions from JSON to Markdown format, fixing an issue where dynamically created subagents failed to invoke.
Fixed a crash occurring when executing background tasks or terminal commands that produce empty outputs (such as sleep).
Fixed shutdown resource leaks by integrating the shared SQLite summary store for background synchronization and resolving goroutine and database connection leaks on CLI exit.
Fixed a permission manager hook error by safely handling empty decision strings returned by pre-tool hooks instead of failing with an "unknown pre-tool hook decision" error.
Subagent & task status bar, editor warnings, and Windows fixes
New interactive status bar for subagents and background tasks, clipboard image pasting improvements, and Windows fixes.
Improvements(5)
Introduced a new interactive status indicator below the input box that displays active subagents and background tasks in real-time, making it easy to monitor and navigate parallel workflows at a glance.
Added ctrl+g on the artifact view to open $EDITOR. Also added a warning confirmation prompt before opening the editor in the artifact detail view if there are unsent comments, and ensured these comments are preserved upon reload if the artifact content was not modified.
Added alt+v as an alternative paste shortcut on Windows to resolve issues where ctrl+v is intercepted by the terminal emulator, enabling reliable image pasting.
Improved the /permissions panel to dynamically reload configurations from disk and prevent accidental overwrites.
Increased the MCP connection timeout to 60 seconds to improve reliability for slow-starting custom MCP servers.
Fixes(4)
Fixed a bug on Windows where print mode and other non-TUI command outputs were silently discarded when run in non-TTY environments (such as pipes or subprocesses).
Fixed Windows editor fallback to use "edit" or "notepad" when the editor setting is "auto" and no editor is configured, instead of attempting to use "vim".
Fixed the subagent approval TUI to dynamically render user-defined custom keybindings (such as alternative approval keys) instead of showing hardcoded defaults.
Fixed alignment and wrapping issues in the comment editor for multiline comments, ensuring all lines are indented consistently.
Optimistic slash commands, strict permission matching, and multiline undo
Optimistic slash command rendering, strict non-regex permission matching, and unified undo/redo prompt history.
Improvements(2)
Improved command permission security by making "Always Approve" rule matching strict (non-regex) by default, while allowing users to explicitly opt-in to regex matching by prepending rules with regex:.
Improved command permission usability by relaxing redirection checks, allowing safe commands with output redirection (e.g., tool > file) to match without requiring strict full-command approval.
Fixes(5)
Fixed a bug where the CLI would temporarily render skill commands without their slash prefix during optimistic updates by deferring prefix stripping to the serialization boundary, ensuring the UI always displays exactly what the user typed.
Fixed a redundant CLI exit message by removing the "Resume in the same project" hint line, leaving only the standard resume command to simplify exit output.
Resolved bugs during UI transitions (such as opening subagent details or logging out) by introducing a unified synchronization mechanism that prevents key lockups and ensures overlay panels like the /help view are properly reset.
Fixed a bug in the CLI prompt editor where undo and redo history stacks could become desynchronized during rapid mutations by decoupling the history state into a unified, pointer-backed structure.
Fixed a bug where browser-related prompt sections were missing from the agent's prompt registry, ensuring browser-based tasks execute reliably.
Added support for --project and --new-project launch flags to allow users to explicitly set or create projects, and updated the project resolution logic to default regardless of the active workspace.
Added a confirmation prompt when pressing Esc in comment mode with unsaved modifications to prevent users from accidentally discarding their work in review views.
Added dynamic OSC8 terminal hyperlink support to render clickable links in supporting terminals, with automatic fallback stripping for backward compatibility.
Introduced reverse diff cycling navigation mapped to shift+n in unified diff review mode to allow users to easily cycle backwards through diff blocks.
Improved permission config merging priorities by ensuring project-specific configurations (located in ~/.gemini/config/projects/) take precedence over global settings in ~/.gemini/antigravity-cli/settings.json.
Fixes(4)
Fixed a regression where ctrl+o scrollback clearing failed by restoring the use of cached fields rather than shared pointer comparisons for trajectory toggle detection.
Fixed a rendering bug where Makefile syntax (like $(call ...)) inside code blocks was mistakenly parsed and mangled by LaTeX math expansion, by introducing a state machine that restricts expansion to prose segments.
Fixed an enterprise network connectivity issue by restoring AES-NI compile-time optimizations, which prevents Deep Packet Inspection (DPI) firewalls from incorrectly flagging and resetting TLS connections.
Fixed incorrect key strings by removing the unsupported backtab default binding and correcting invalid pgdn references to pgdown to align with Bubble Tea v2 canonical names.
Ctrl+C interrupt flow, persistent resume cache, and AltScreen tool review
Added ctrl+c interrupt and exit handling, parallel /resume metadata caching, and expanded AltScreen tool confirmations.
Improvements(12)
Added ctrl+c as an exit and interrupt key: the first press cancels active agent operations (like streaming responses), and a double-press triggers the exit flow. Also added a dynamic exit hint in the status line.
Improved /resume loading performance by implementing a persistent metadata cache and parallel loader, eliminating severe latency with large conversation histories and preventing background loading log spam.
Added an expanded AltScreen view for tool confirmations (accessible via ctrl+g), allowing users to view and edit the full command and associated permissions in a dedicated full-screen view, replacing the inline edit (e) key.
Added the AGY_CLI_CMD_OUTPUT_PERCENTAGE environment variable, allowing users to customize the maximum height of command outputs in the TUI as a percentage of the terminal height.
Added strict key name validation to the keybindings system to reject invalid key names (like typos) and suggest canonical alternatives, preventing "dead keys" from being registered.
Added a validation warning when ctrl+c is mapped to a non-default action, clarifying that the system always intercepts ctrl+c to interrupt active operations or exit, and providing instructions on how to resolve the warning.
Improved command output rendering by making the output height dynamic, improving the readability of commands like /keybindings.
Improved text rendering with ANSI-aware word wrapping at word boundaries and prevented URLs containing hyphens from being incorrectly split across lines.
Improved the /resume experience: added support for pasting clipboard text into the search filter and rename fields, upgraded the rename input to a multiline editor to prevent long titles from being hidden, and fixed a bug where the navigation cursor could disappear.
Improved keybinding validation warning messages to use user-facing names (e.g., cli.escape) instead of internal representation names.
Improved startup behavior by only creating the keybindings.json configuration file when the user explicitly runs the /keybindings customization command, rather than automatically generating it on every startup.
Improved keybinding error presentation by replacing the persistent error footer with transient error alerts, freeing up valuable terminal space.
Fixes(4)
Fixed ctrl+d behavior to act as a forward-delete when the input prompt contains text, only triggering the exit flow when the prompt is empty.
Fixed the ctrl+c exit safety valve to ensure it always works as an interrupt or exit key, regardless of how it is mapped in the user's custom keybindings configuration.
Fixed VCS commit tree rendering to reserve the @ marker exclusively for the actual current commit in the VCS history rather than the synthetic "Working Copy" entry, helping users easily identify the working copy parent.
Fixed authentication error handling to gracefully handle unsigned-in states by returning an empty configuration and suppressing noisy error logs.
Git submodule plugins, read-only permissions, and sandbox hardening
Automatic Git submodule resolution for plugins, read-only access for builtin skills, and sandbox path hardening.
Improvements(4)
Added submodule support for plugins installation. External plugin installation now automatically resolves and initializes Git submodules.
Optimized customizations permissions: Automatically grants read-only access to the builtin customizations directory, eliminating redundant permission prompts on startup.
Improved glamour parser error handling (like nested checkboxes inside list emphasis) and preventing it from crashing the TUI, falling back to raw text with a warning banner.
Updated bubbletea to v2.0.7: Resolves a potential TUI panic when terminal input is unavailable, fixes a data race in mouse handling within the Cursed Renderer, and corrects mouse release behavior under the Kitty Keyboard protocol.
Fixes(5)
Hardened command execution permission checks by enforcing strict exact-match verification for PowerShell scripts, complex shell redirections ( > , 2>&1 ), and unparseable strings to prevent sandbox escapes.
Hardened sandbox execution by adding .git to the core list of dangerous paths, preventing unauthorized or destructive repository modifications.
Fixed a bug where allowlisted terminal commands with quoted arguments (e.g., python -c "print(1)") would silently fail to match at runtime due to flawed whitespace tokenization.
Fixed a bug in headless print mode resumption (--conversation/-c-p ...) where the CLI would dump the entire historical conversation transcript instead of only printing the newly generated response.
Fixed a CPU compatibility issue on ARM64 devices without AES hardware support.
Slash command history, Models & Quota page, and TUI performance
Replay slash command history with Up arrow, redesigned Models & Quota page, and statusline quota usage indicators.
Improvements(9)
Added support for capturing slash command history, allowing users to use the up arrow to replay previously entered slash commands.
Redesigned the "Models & Quota" page (enabled by default, replacing the legacy usage page) to gracefully handle disabled quota buckets by displaying a dimmed "Disabled" status and omitting the progress bar.
Added display of quota usage and execution mode in the status line.
Improved /btw to be more token efficient and support streaming responses for a smoother user experience and fixed premature truncation.
Added a per-line guard against extremely long single-line pastes in the TUI prompt editor to prevent performance lag, replacing them with an expandable placeholder.
Redesigned the /resume conversation picker to align the workspace column and added adaptive column dropping (workspace, time, steps) to support narrow terminals.
Redesigned the /tasks list and detail views for better alignment and readability, placing start times on the left, right-aligning status, and capping the panel height.
Improved configuration saving by propagating write failures as transient error flashes on the statusline.
Improved settings inheritance by ensuring the CLI inherits the use_ai_credits setting from global user settings on startup.
Fixes(8)
Fixed a bug where the /hooks command wrote configurations to ~/.gemini/antigravity-cli/hooks.json instead of the shared ~/.gemini/config/hooks.json, ensuring hooks remain synchronized between the TUI and the backend.
Fixed a CPU compatibility issue (SIGILL on non-AES-NI CPUs), preventing immediate crashes on startup on older CPUs (like Intel Ivy Bridge) or VM environments that lack AES-NI support.
Fixed dynamic reloading of custom skills and system slash commands, ensuring they are instantly discovered in autocomplete upon conversation switch or /add-dir.
Fixed a TUI hang in the artifact view during long sessions by optimizing the rendering complexity of large step histories.
Fixed an autocomplete bug where a command that is an exact prefix of another (e.g., /conv vs /conv-switch) would aggressively auto-complete and hide the suggestions menu.
Fixed a race condition where sending a message immediately after denying a permission request would fail due to incomplete backend cleanup.
Fixed potential OOM risks when reading large clipboard files by verifying file size before reading.
Fixed Windows and Wayland-only Linux distributions clipboard image and file reading.
Configurable MCP timeouts, artifact gutter line mapping, and Wayland clipboard
Configurable MCP server connection timeouts, accurate line-numbered artifact gutters, and native Wayland clipboard support.
Improvements(6)
Added a configurable timeout for launching MCP servers, allowing users to specify a custom timeout or set it to -1 to disable the timeout completely.
Revamped the artifact viewer gutter numbering and line mapping to accurately align terminal viewport lines with actual 1-based source file line numbers, including support for wrapped lines and collapsed Mermaid diagrams.
Increased the maximum tool calls limit to 512 for Gemini models, allowing agents to perform significantly more complex, multi-step tasks in a single turn.
Added support for installing plugins directly from GitHub subpaths (with branch resolution).
Added native Wayland clipboard support (wl-paste) on Linux, falling back to xclip for X11 environments, and prioritized copied files (from file managers) over raw image data.
Preserved unknown fields in settings.json during read, write, and merge operations, preventing settings from being silently wiped out when switching between different CLI versions or builds.
Fixes(8)
Fixed a bug where the CLI could get stuck in a pending state (showing a transient spinner) after sending a message due to stale status updates.
Fixed a bug where the wrong workspace directory was displayed in the header and /help menu when multiple workspaces were active.
Fixed a desync bug in the agent state management where stale callbacks from previous runs could be used upon cache hits in new agent state.
Fixed Windows-specific sandbox network proxy issues, resolving a hang during connection hijacking and correcting tunnel response protocols.
Fixed a bug where the archival status timestamp was not correctly saved when archiving conversations.
Fixed a potential stack overflow crash by introducing a non-recursive warning output mechanism for pre-conversation errors.
Fixed variable resolution in plugins, ensuring gemini cli variables like ${extensionPath} correctly resolves to the final installation directory.
Fixed layout boundary overflow, scrolling visibility, and out-of-bounds scrolling bugs in the artifact detail view when inline comments are present.
Path auto-completion, optimistic prompt rendering, and fuzzy matching
Path autocompletion for /open and /add-dir, optimistic chat prompt rendering, and fuzzy slash command matching.
Improvements(5)
Added shell-style path auto-completion for /open and /add-dir.
Added optimistic rendering for user chat prompt submissions, injecting messages immediately into the viewport to eliminate perceived input lag.
Added fuzzy and partial substring matching across slash commands. E.g. /el -> shows /help and /model while previous no suggested completions.
Skipped subagent conversations from /resume, keeping the standalone conversation picker focused purely on direct user initiated conversations.
Added a stack_with_default flag to the statusLine configuration to render both the default Antigravity status line and custom status line output vertically stacked.
Fixes(3)
Fixed a bug when suggestion was not triggered when @ is typed after (. Enabled unconditional typeahead suggestions whenever @ is typed without preceding whitespace, streamlining mention workflows.
Fixed a bug where entering a prompt immediately after pressing Esc (to interrupt an active agent stream) caused the newly typed input to be swallowed or rejected.
Fixed --sandbox flag propagation in headless print mode (-p / --print), ensuring sandbox isolation is correctly enforced during non-interactive execution.
Added --model to set model when launching CLI. Also a new models subcommand to list available models.
Added /permissions command which allows to add/edit/remove permissions rules for each of the three configs above directly inside the CLI.
Allowed opening the Artifact Review panel (shortcut ctrl+r) while answering pending questions or tool permission confirmations, preserving your current progress when toggling back.
Improved statusline layout by merging active tip and artifact status on a single line and truncating with ellipsis on narrow terminals to prevent collisions.
Improved customization support by allowing directories in the customization manager to be passed as workspace directories, enabling correct trajectory metadata population and /add-dir support.
Added support for url in mcp_config.json to configure MCP servers directly via a URL.
Improved /resume performance: optimized lazy loading of conversation details, filtered out empty conversations, and added support for scanning SQLite database files (.db and .db-wal).
Improved autocomplete: tab completion for slash commands now resolves to the matched alias instead of the primary command name (e.g., /se autocompletes to /settings instead of /config).
Integrated the permissioning system with the rest of Antigravity. CLI permissions now merges project level permissions, permissions from user settings shared with Antigravity, and permissions from the CLI settings.json.
Fixes(1)
Fixed a bug that metadata was written in the current directory as opposed to ~/.gemini/antigravity-cli/cache when running using -p.
SQLite conversation store, LaTeX math rendering, and centralized project cache
SQLite conversation database format, terminal LaTeX math rendering, and centralized project discovery cache.
Improvements(8)
Added SQLite (.db) conversation support and will be CLI’s conversation format. Fixed a bug when importing SQLite conversation from Antigravity 2.0 to CLI.
Added LaTeX math rendering, enabling the CLI to display beautiful mathematical formulas directly in the terminal viewport. Set AGY_CLI_DISABLE_LATEX environment variable to turn off LaTeX rendering globally if desired.
Decoupled project discovery from local .antigravitycli workspace directories. The CLI now stores workspace-to-project mappings in a centralized ~/.gemini/antigravity-cli/cache/projects.json file, eliminating repository clutter and speeding up project discovery to a single-map lookup.
Collapses all newlines and consecutive whitespaces in conversation previews and titles before rendering list items, preventing visual UI layout breaks in the picker rows.
Styled the separator space between the line number column and diff content to match the text blocks, ensuring background highlights stretch seamlessly across the viewport width in tool outputs and /diff details.
Aligned the interactive /changelog and agy changelog cache paths to both use antigravity-cli, and made the caching process synchronous to resolve a race condition where immediate process exit terminated the cache write.
Moved VCS detection out of the synchronous CLI startup path to prevent slow initialization.
Parallelized the MCP server initialization sequence, preventing slow or hanging custom MCP servers from blocking independent, fast-starting servers (like local plugins) from loading on startup or configuration reloads.
Fixes(3)
Resolved sporadic and permanent UI hangs caused by a stateful callback streamer race condition during network drops or extremely fast agent steps.
Resolved inconsistent behavior where selecting skill-derived slash commands from autocompletion suggestions cleared the input without executing. Autocompleted skill commands are now correctly submitted to the backend.
Resolved an issue where exclusion rules and allowlists configured in rules.json were silently ignored, causing the discovery engine to load every .md rule file unconditionally at boot.
G1 AI credits support, /credits panel, and input prompt fixes
G1 AI credits support when standard quotas run out, in-CLI /credits panel, and prompt input fixes.
Improvements(4)
Added support for G1 credits in the Antigravity CLI. Users can now utilize G1 credits when their standard quota runs out. This includes a new UseG1Credits setting to enable automatic credit usage and a real-time display of remaining credits in the status bar.
Added a new /credits panel that provides an in-CLI interface with a direct link to purchase additional G1 credits.
Redesign CLI logo on Apple Terminal.
Improved color scheme preview in settings and onboarding: added warnings and thought process examples to the preview, and corrected link styling to only underline the URL.
Fixes(6)
Fixed an infinite loop in the prompt input. Navigating left (wordLeft) when encountering spaces at the very beginning of the input no longer causes an infinite hang.
Fixed custom MCP server disabling via the TUI. Resolved a directory path mismatch where pressing the [Disable] button wrote to the legacy mcp_config.json path instead of the migrated config/mcp_config.json path, ensuring custom MCP servers can now be successfully disabled and unloaded.
Fixed $EDITOR environment variable parsing: resolved issues where arguments containing = (e.g., --alternate-editor=vi) were incorrectly split, causing editor launch failures.
Fixed /diff detail view truncation: implemented dynamic line wrapping based on terminal viewport width and added automatic tab-to-space expansion to prevent layout overflow.
Fixed project discovery robustness: updated the CLI to skip invalid or broken symlinks in .antigravitycli/ rather than failing immediately, allowing discovery of valid projects.
Fixed AskQuestion state management: memorizes selected options, write-in values, and UI states when navigating back and forth (KeyLeft) between questions in multi-question dialogs.
Account hiding flag, subagent timeouts, and TUI fixes
Added AGY_CLI_HIDE_ACCOUNT_INFO flag, subagent-specific interaction timeouts, and TUI fixes.
Improvements(2)
Added AGY_CLI_HIDE_ACCOUNT_INFO environment variable to hide email and plan tier from the header.
Improved /help shortcuts tab by sorting shortcuts by keybinding key, adding missing keybindings (like ctrl+r, ctrl+o, alt+j, ctrl+k), and generalizing scrolling (PageUp/PageDown/GoToTop/GoToBottom) for both Commands and Shortcuts tabs.
Fixes(7)
Fixed timeout overrides: restricted the default 60-second interaction timeout specifically to subagents, preventing the main agent from being unconditionally capped.
Fixed a nil-pointer panic in Sandbox Mode: resolved a typed nil interface comparison when fetching URL content.
Fixed fallback skill discovery in Standalone mode: ensures custom/fallback skills are successfully loaded even if the standard configuration directory is missing, and added automatic path deduplication to prevent duplicates.
Fixed command rendering in message history: prefixed slash commands with a caret (>) in response block headers to clearly distinguish user-typed commands from agent outputs.
Fixed plugin installation path mismatch: updated the plugin subcommand to install downloaded plugins directly to the shared configuration directory (~/.gemini/config/) rather than the private application data folder, making them instantly discoverable.
Fixed Git short-hash support in diff selection: updated the commit hash recognition pattern in the /diff commit selection tree to match Git's standard 7-character short hashes (and up to 40-character full hashes).
Fixed statusline subcommand handling and recursive loops: added case-insensitive subcommand parsing (help, delete, reset, enable/on, disable/off) to the /statusline command, providing direct control to toggle or revert custom statuslines and blocking recursive shell hangs during help queries.
Added proceed-in-sandbox tool permission mode. Auto-approves terminal commands that run inside the secure sandbox, requesting manual approval only when a command attempts to bypass the sandbox.
Integrates consumer/free-tier onboarding directly into the CLI.
Added plugin discovery for skills and agents. Automatically scans installed plugin directories to make custom skills and specialized agents available for execution in the CLI.
Moves the terminal color scheme to the top of the selection list, making it the default choice during onboarding and in /settings.
Improved /usage and /quota commands. Forces a real-time reload of model configuration and remaining quotas, allowing you to see updated real-time consumption statistics immediately.
Improved step rendering layout. Calculates available terminal width dynamically and uses middle-truncation (/foo/.../bar) for file path tools to prevent layout shifting on narrow screens.
Improved session deletion keybinding in /resume. Changed the shortcut from ctrl+d to ctrl+delete to resolve conflicts with the global exit keybinding (ctrl+dctrl+d) and preserve Emacs-style forward-delete in search input fields.
Restored automatic table wrapping, preventing long cells inside markdown tables from being truncated.
Fixes(5)
Fixed OAuth token persistence and authentication hangs.
Fixed Windows log redirection and resizing issues. Resolved a critical bug where logs were not redirected correctly on Windows, which previously caused the terminal to swallow window resize events and shut down slowly.
Fixed pasted text line counting. Corrected line counting for user inputs to ensure extremely long inputs are correctly folded into a [Pasted text #X +Y lines] placeholder to keep the viewport clean.
Fixed onboarding stability. Resolved a race condition where a concurrent terminal resize event during onboarding could revert the UI to a blank onboarding screen.
Resolved an issue where deleted files (represented by +++ /dev/null) had their deletion lines incorrectly merged into the previous file's diff.
Local model support with LiteRT and LocalOpenAI, standardized evaluation presets, and custom subagent models
This release announces official Antigravity SDK local model support with first-class LiteRTAgentConfig and LocalOpenAIAgentConfig configurations, introduces a standardized evaluation preset (AgentConfig.eval()) for core coding evaluations, enables custom subagent model overrides, turns on the schedule tool by default across standard tool groups, and pairs background task management with command execution.
Improvements(13)
Antigravity SDK + Local Models Support: Official local model support is now ready to use with first-class LiteRTAgentConfig and LocalOpenAIAgentConfig configurations. Developers can execute on-device models with automated lightweight presets or integrate with local OpenAI-compatible endpoints.
Standardized Evaluation Preset (AgentConfig.eval()): Adds a standardized, benchmark-ready preset that configures autonomous permissions, benchmark retry behavior, daemon execution in commands, and strips image generation and subagents to focus on core coding evaluations.
Custom Subagent Models: Allows developers to assign specific model targets to subagents independently of the root agent configuration.
Built-in Schedule Tool: Enables the schedule tool by default across standard tool groups, permitting agents and subagents to set timers and cron-based background jobs alongside task management.
Automatic Lightweight Presets for LiteRT: Instantiating LiteRTAgentConfig now automatically applies optimized lightweight presets—including reduced prompt overhead and synchronous context compaction—without requiring an explicit call to .lightweight().
Task Management Pairing: Automatically enables the manage_task tool whenever run_command or schedule is active, allowing background task lifecycle management.
LiteRT & Local Model Examples: Added getting-started guides and end-to-end examples demonstrating on-device execution with LiteRTAgentConfig (Gemma 4 26B) and OpenAI-compatible local endpoints via LocalOpenAIAgentConfig.
Sandbox Availability Warning: Added a session startup warning when enable_sandbox=True is requested on an environment or OS backend where sandbox isolation cannot be enforced.
Extended JSON Schema Normalization: Added schema normalization support for OpenAPI and JSON Schema Draft 7 / 2020-12 keywords (such as multipleOf, prefixItems, and dependentSchemas) and prevented accidental mutation of uppercase sample values.
Optimized ToolRunner Coercion: Improved type resolution for closure-scoped tools and forward-referenced ToolContext parameters, and introduced TypeAdapter caching to accelerate tool call execution.
Top-Level ServiceTier Export: Re-exported ServiceTier at the root package namespace for easier import parity.
Excluded ASK_QUESTION from Default Tools: Excluded BuiltinTools.ASK_QUESTION from BuiltinTools.default(). Default configurations run autonomously; agents in headless workflows will no longer attempt interactive user prompts. To re-enable interactive questions, explicitly pass BuiltinTools.ASK_QUESTION in enabled_tools.
Single Compaction Threshold Dial: Simplified CompactionConfig to a single token_threshold property, deprecating legacy context token limits and interval dials. To configure context compaction, specify CompactionConfig(token_threshold=...).
Fixes(1)
Service Tier Ingestion: Fixed an unhandled ValueError when connecting via gateways reporting unlisted backend service tiers (such as Vertex AI PROVISIONED_THROUGHPUT) by safely ignoring unknown tier values while preserving token counts.
Patches(1)
Removed Deprecated modified_arguments_json: Removed the legacy JSON string fallback in tool hook interception in favor of modified_args.
This release introduces first-class conversation compaction controls via CompactionConfig, delta and forward-looking budget scopes for session resumption, tool output token truncation limits, and expanded arithmetic operations on UsageMetadata. It also delivers OS-level terminal sandboxing examples and automated tool wrapper reflection preservation.
Improvements(7)
Compaction Configuration: Adds CompactionConfig on AgentConfig to govern sliding-window conversation history compaction through an explicit token ceiling dial (token_threshold), deprecating CapabilitiesConfig.compaction_threshold.
Forward-Looking Budget Scope: Adds BudgetScope.FORWARD_LOOKING to BudgetConfig to enforce model call and token budgets across newly resumed execution turns without counting previous historical usage.
Tool Output Token Truncation: Exposes tool_output_truncation_config across agent configurations to cap token output volume initially for run_command executions in localharness.
UsageMetadata Arithmetic: Supports standard Python arithmetic protocols on UsageMetadata, enabling scalar multiplications, scaling, and accumulating token counts using built-in sum().
Tool Wrapper Metadata Preservation: Preserves function signature, module name, annotations, and original callable references via __wrapped__ when registering tools with ToolWithSchema.
Interactive REPL Policy Flattening: Automatically flattens nested policy lists during interactive REPL upgrades so nested command authorization rules correctly upgrade to prompt the user.
Terminal Command Sandboxing Guide: Added getting-started guide and references demonstrating OS-level command sandboxing using RunCommandConfig(enable_sandbox=True) paired with execution policies.
Fixes(3)
Tool Call Deserialization: Fixed dropped tool arguments during tool call handling when incoming payloads provide structured dictionary arguments rather than serialized JSON strings.
Local Step Trajectory Tracking: Fixed missing provenance metadata by forwarding trajectory_id from incoming tool calls to local connection execution steps.
Dynamic Content Proto Resolution: Fixed an AttributeError when importing struct_converter in external environments missing internal protobuf definitions by resolving descriptor types dynamically at runtime.
Default Gemini 3.8 Flash model upgrade, lightweight agent configuration, and Vertex AI Express mode
The 0.1.16 release updates the default model for new agents to gemini-3.8-flash for enhanced reasoning capabilities, introduces a .lightweight() method on agent configs for small and local models, adds Vertex AI Express mode with API keys, and expands tool runner support for callable classes and dataclass functors.
Improvements(8)
Default Model Update to Gemini 3.8 Flash: The default model for new agents and configurations has been updated to gemini-3.8-flash, delivering higher reasoning quality and stronger task performance. Developers can override this default by specifying model in their configuration:
Optimized Lightweight Agent Configuration: New .lightweight() method on agent configurations (including LocalAgentConfig and LiteRTAgentConfig) applies preset optimizations for smaller, local-running models by configuring minimal tool sets, minimal prompting, and disabling background subagents:
Vertex AI Express Mode (API Key Support): Developers can now connect to Vertex AI using Express mode by providing an API key directly on LocalAgentConfig(vertex=True, api_key="..."), simplifying authentication without needing full GCP project/location ADC setup:
Support for Callable Class and Dataclass Tools: The SDK's ToolRunner now inspects and executes tools defined as callable class instances (functors) and dataclass methods, enhancing flexibility for custom tool implementations:
OS Sandbox Opt-in for Commands: Added an enable_sandbox field to RunCommandConfig to allow developers to execute terminal commands within an OS-level sandbox environment:
Tool Invocation Argument Flexibility: ToolWithSchema and public callable proxies now accept both positional (args) and keyword arguments (*kwargs), matching standard Python function calling conventions.
Support for genai.Content Media: Introduced a structconverter to support media blocks from genai.Content objects within the SDK, enabling richer multimodal interactions.
MCP Dependency Compatibility Widening: The SDK now supports both mcp>=1.0 and mcp<3.0 dependencies.
Fixes(3)
Stop Hook Integration: Agents now support the StopHook lifecycle hook, triggered when an agent's execution is externally stopped, enabling custom cleanup and resource logging actions:
Interactive Example Agent Behavior: Fixed interactive SDK examples (interactive_cli.py, human_in_the_loop.py) to consistently configure AgentBehavior.INTERACTIVE.
Tool Runner stdio MCP Environment: Resolved an issue in exported SDK examples where stdio MCP servers failed by ensuring child processes use sys.executable within the active virtual environment.
Patches(1)
Step Token Usage Deprecation: Deprecated Step.usage_metadata in favor of consolidated turn-level (ChatResponse.usage_metadata) and session-level (agent.conversation.total_usage) reporting.
The 0.1.15 release introduces subagent-exclusive tool scoping to reduce context overhead, adds an on_compaction lifecycle hook for observing context checkpointing, expands platform compatibility with Alpine Linux musl wheels, resolves workspace path normalization and custom Vertex endpoint routing, and adds universal JSON Schema normalization for custom Python tools when targeting local OpenAI-compatible LLM endpoints.
Improvements(9)
Subagent-Scoped Custom Tools: Custom tools can now be registered directly on subagents without requiring registration on the root agent, strictly isolating them to the subagent's context and reducing root context tokens.
Context Compaction Lifecycle Hook: Improved support for the on_compaction lifecycle hook to accurately capture compaction events and summaries when long-running conversations trigger checkpointing.
Custom Base URL & Secure Vertex Proxy Support: VertexEndpoint now supports routing requests to custom base_url reverse proxies or enterprise gateways without leaking ambient Google Cloud Application Default Credentials (ADC) OAuth tokens or conflicting with project/location configurations.
Universal JSON Schema Normalization for Custom Tools: Custom Python tools now produce standard OpenAPI / JSON Schema compliant parameter definitions with lowercase types and camelCase combiners, eliminating schema errors when connecting to local OpenAI-compatible engines such as Ollama, LM Studio, or vLLM.
from_bytes Helper Export: Exported the from_bytes helper in top-level google.antigravity alongside from_file for creating binary and multimodal content payloads.
PEP 656 musllinux Wheel Support: Added musllinux_1_1 wheel platform tags for x86_64 and aarch64 architectures, enabling direct installation via pip on Alpine Linux containers.
Isolated Harness Environment Configuration: Added per-connection environment dictionary resolution for ANTIGRAVITY_HARNESS_PATH, avoiding mutation of global os.environ.
Tool Call Metadata Preservation: Preserved id, step_id, and server_name metadata across all ToolResult executions, including batch calls, errors, and unknown tools.
Relative Workspace Path Normalization: LocalAgentConfig(workspaces=...) now resolves relative directory paths and tilde (~) expansions against the current working directory (os.getcwd()).
Fixes(6)
Local OpenAI Tool Schema Validation: Fixed HTTP 400 "Invalid discriminator value" errors on local OpenAI endpoints by canonicalizing tool parameter schemas to standard JSON Schema.
Vertex Custom Gateway Token Leaks: Fixed host GCP OAuth bearer token leakage and environment variable collisions when using custom base_url endpoints with VertexEndpoint.
Subagent Custom Tool Routing: Fixed tool dispatch failures when subagents defined tools not registered on the root agent.
UsageMetadata Service Tier Loss: Fixed UsageMetadata.__sub__ dropping the service_tier field when calculating per-turn usage deltas.
WebSocket Deprecation Warnings: Resolved DeprecationWarning exceptions when reading WebSocket close codes across varying websockets library versions.
OpenTelemetry Optional Dependency Guard: Fixed test collection failures on minimal environments lacking opentelemetry.sdk.
Patches(1)
Deprecated TriggerDelivery Removal: Removed the unused TriggerDelivery enum from types.py.
Workspace path resolution against current working directory, compaction event deduplication, and Vertex auth overrides
Resolves relative workspace paths against the current working directory, prevents duplicate compaction hook dispatches and compaction index inflation, and adds support for Vertex AI gateway authentication overrides.
Improvements(1)
Vertex Endpoint Authentication Overrides: Enhanced VertexEndpoint to support custom authentication headers and token overrides when routing inference requests through API gateways or proxy servers.
Fixes(2)
Workspace Path Resolution: Fixed relative workspace paths in LocalAgentConfig(workspaces=[...]) incorrectly resolving against app_data_dir instead of the process working directory.
Compaction Hook Deduplication: Fixed on_compaction hook dispatching multiple times per compaction event and resolved over-counting in compaction_indices by adding terminal state deduplication.
The 0.1.13 release introduces pre-tool argument modification capabilities in lifecycle hooks, adds native support for synchronous hook functions, and establishes structured command execution configuration with configurable execution timeouts. It also improves tool execution observability with step correlation IDs and enhances connection resilience during client disconnects.
Improvements(5)
Pre-Tool Hook Argument Modification: Pre-tool lifecycle hooks can now sanitize, transform, or override tool input arguments before tool execution begins.
Synchronous Hook Function Support: Lifecycle hook decorators now accept standard synchronous functions alongside asynchronous coroutines without raising runtime await errors.
Structured Command Execution Configuration: Command execution settings are now consolidated under RunCommandConfig, introducing configurable timeouts that default to 10 minutes (600 seconds) alongside daemon execution controls.
Tool Lifecycle Step Correlation: ToolResult and ToolExecutionError event payloads now include step_id, enabling end-to-end tracking and correlation of tool invocations across trajectory steps.
VS Code Debugging Configuration: Updated setup_vscode_debugging.sh to target the canonical getting_started/hello_world starter example and explicitly configure Gemini Developer API defaults.
Fixes(2)
Synchronous Hook Decorator Execution: Fixed a runtime TypeError when decorating synchronous functions with @pre_turn, @post_tool_call, and other lifecycle hooks by verifying awaitability before awaiting hook responses.
LiteRT Early Client Disconnects: Suppressed unhandled ConnectionResetError, ConnectionAbortedError, and BrokenPipeError exceptions when clients disconnect early from local LiteRT server connections.
Fix proto deserialization regression for ActionGenerateImage output paths
The 0.1.12 release fixes a client-side proto deserialization regression in image generation result handling.
Fixes(1)
ActionGenerateImage Proto Skew: Resolved client-side deserialization errors where the bundled binary emitted ActionGenerateImage.output_path while the shipped Python proto lacked the corresponding field, ensuring robust protobuf parsing across image generation steps.
Default Gemini 3.7 Flash upgrade, session budget limits, Vertex AI Express Mode, and autonomous behavior modes
The 0.1.11 release updates the default model to gemini-3.7-flash, introduces session-level budget enforcement and turn termination stop reasons, Vertex AI Express Mode authentication, and an autonomous agent behavior setting. It also expands tool hook metadata, resolves string annotation coercion for postponed evaluation, and improves MCP server and subagent stability.
Improvements(6)
Default Model Upgrade to Gemini 3.7 Flash: Upgraded the default inference model to gemini-3.7-flash.
Session Budget Enforcement & Stop Reasons: Added BudgetConfig to define session-level usage limits (total tokens, turns, and cost) and StopReason enum (BUDGET_EXCEEDED, TURN_LIMIT, USER_CANCELLED, etc.) to inspect turn termination causes.
Vertex AI Express Mode Support: Added native support for Express Mode authentication via VertexEndpoint(api_key=...) and LocalAgentConfig(vertex=true, api_key=...), simplifying headless and non-GCP deployments.
Autonomous Agent Behavior Mode: Control the agent's behavior with AgentBehavior. By default the SDK now has an AgentBehavior.AUTONOMOUS mode (previously AgentBehavior.INTERACTIVE) to streamline scripting, background, and headless interaction modes. Override by setting CapabilitiesConfig(agent_behavior=AgentBehavior.INTERACTIVE).
Multi-Interface Hook Registration: Enabled single-instance registration across multiple hook interfaces (PreToolHook, PostToolHook, PreTurnHook), allowing for cross-functional instrumentation without duplicate invocations.
PreToolArgs Metadata: Exposed trajectory_id and step_index on tool hook payloads for chat thread context tracing.
Fixes(4)
ToolRunner String Annotation Coercion: When using from __future__ import annotations, tool argument coercion failed on stringified types; resolved by resolving type annotations via typing.get_type_hints before type adaptation.
MCP Server Example Port Binding: Ephemeral port race conditions in test and example server startup were resolved by binding directly to port 0.
Subagent Deadlock Prevention: Handled subagent fatal errors in the localagent executor to prevent deadlocks when subagents terminate abnormally.
Empty Input Validation: Added input validation to prevent SDK unresponsiveness on empty or whitespace-only prompts.
The 0.1.10 release introduces support for Gemini Prioritized Inference service tiers, real-time token usage event streaming, stateful context-aware hook decorators, and explicit tool call correlation IDs across hook callbacks. It also standardizes default system instruction merging behavior, fixes WebSocket compaction events, and expands local Gemma model documentation.
Improvements(10)
Gemini Prioritized Inference Service Tier: Configure agents to utilize Gemini Prioritized Inference service tiers for high-priority model execution with automated graceful fallback.
Tool Call ID Correlation in Lifecycle Hooks: Inspect call_id attributes on tool executions, errors, and hooks to correlate multi-step tool invocations across lifecycle callbacks.
Context-Aware Hook Decorators: Decorate hook handlers (@hooks.pre_turn, @hooks.post_tool_call, etc.) that optionally accept HookContext as a parameter to maintain state and share data across lifecycle callbacks.
ActionCompaction Event Emission & Hook: Track context window compaction notifications over WebSockets and intercept them using @hooks.on_compaction.
Standardized System Instructions Strategy: Plain string instructions default to appending to built-in instructions. To override and completely replace built-in instructions, pass CustomSystemInstructions.
Live Token Usage Reporting: Introduced real-time UsageUpdate event streaming so token usage accumulates live during agent execution rather than delaying updates until state transitions.
Interactive CLI Spinner: Updated CLI interactive loop spinner to list all active tool names when running concurrent tool calls (e.g., Running tools 'tool_a', 'tool_b').
Module Re-exports: Re-exported ReadUrlContentResult and SearchWebResult in connections.local for uniform tool result access.
Local Gemma Model Documentation: Added guides and tutorials for running agents locally with Gemma models using LiteRT and OpenAI-compatible endpoints.
LiteRT Token Output Limit: Increased max_output_tokens default in LiteRT local server configuration from 8,192 to 16,384 tokens to prevent truncation during complex reasoning and generation tasks.
Fixes(2)
ActionCompaction Event Emission: Fixed issue where compaction notifications were suppressed in external SDK releases, causing @hooks.on_compaction handlers and conversation.compaction_indices tracking to fail; compaction events now emit properly over WebSockets.
MCP Test Server Startup: Fixed a race condition where the HTTP port was exposed before uvicorn server startup completed, which previously caused intermittent ConnectionRefusedError failures during test initialization.
Release 0.1.9 of the Google Antigravity Python SDK adds model-call retry/backoff configuration, improves tool registration for custom functions, adds missing BuiltinTools exports, fixes user audio payload processing, and adds connection DebugConfig support along with ToolExecutionError handling.
Improvements(5)
Model-Call Retry & Backoff Configuration: Exposes configurable retry and backoff parameters in LocalAgentConfig for model calls.
LiteRT Warm-up Timeout Scaling: Dynamically scales LiteRT engine warm-up timeouts based on context size and synchronizes warm-up cleanup.
Improved Tool Registration: Enhances automatic tool registration and docstring parsing for custom Python functions.
BuiltinTools Typing Exports: Exports BuiltinTools at top-level package and typing boundaries for subagent tool configuration.
Connection DebugConfig: Adds base DebugConfig options for enhanced connection debugging and logging.
Fixes(2)
Audio Payload Processing: Fixes processing for user audio payloads in interactive agent sessions.
ToolExecutionError Handling: Introduces ToolExecutionError to SDK error types for explicit tool failure handling.
Release 0.1.8 of the Google Antigravity Python SDK updates the default text model to Gemini 3.6 Flash, adds support for custom subagent instructions, and implements automatic Pydantic argument coercion for tool calls. It also brings defensive prompt sanitization, configurable tool retries, and comprehensive stability improvements for local agent execution.
Improvements(7)
Default Model Upgrade to Gemini 3.6 Flash: Upgrades the default generative text model in the Python SDK to gemini-3.6-flash.
Pydantic TypeAdapter Tool Argument Coercion: Automatically coerces stringified numbers, booleans, and nested models returned by LLMs into strict Python types declared in tool signatures.
Custom Subagent Instructions: Adds capability to specify custom system instructions and allowlisted tool configurations for spawned subagents in multi-agent workflows.
Prompt Sanitization: Strips null bytes (\x00) and non-printable control characters (DEL, BEL, C1) from incoming user prompts at the wire boundary to prevent HTTP 400 errors and terminal corruption.
Configurable Tool Retries: Add RetryConfig definitions in local configuration to allow fine-tuned model and output retry policies.
Operating System Telemetry: Populate client OS and version information in telemetry headers to improve diagnostic tracking.
Native Usage Accumulation: Enable native addition (+ and +=) operations on UsageMetadata instances.
Fixes(7)
Large Tool Output Handling: Dynamic output truncation and removal of WebSocket frame limits to resolve connection termination on large tool responses.
Subagent Idle Synchronization: Fix race conditions where multiple idle states could cause receive_steps() to hang indefinitely.
Multi-threaded state handling, subprocess isolation, extra high thinking, and local model MCP
Release 0.1.7 of the Google Antigravity Python SDK expands end-user control over agent execution environments, strengthens concurrency safety for stateful tools, and introduces deeper reasoning capabilities. Key highlights include atomic multi-threaded state handling for tools and hooks, customizable subprocess environment variable isolation, support for an "extra_high" thinking severity level, and full Model Context Protocol (MCP) and subagent support across local model backends. This release also resolves interactive console prompt clobbering and improves socket discovery under containerized setups.
Improvements(9)
Multi-threaded Hook & Tool State Handling: Developers can now safely read and mutate shared context variables across multi-threaded tools (asyncio.to_thread or ThreadPoolExecutor) using atomic updates and thread locking:
Custom Subprocess Environment Variables: Developers can now pass custom environment variables directly to isolated agent instances via LocalAgentConfig, avoiding pollution of the global parent environment:
Extra High Thinking Severity Support: Developers can now configure an "extra_high" thinking severity level for complex reasoning tasks without needing to specify or override the model name:
MCP & Subagent Support for Local Models (LiteRT & OpenAI): Developers using local Gemma (LiteRTAgentConfig) or local OpenAI-compatible endpoints (LocalOpenAIAgentConfig, such as Ollama or LM Studio) can now directly configure subagents and register Model Context Protocol (MCP) servers, enabling full local multi-agent and MCP tool workflows.
Environment Hydration: Hydrates GCP/Vertex parameters (project, location, routing) dynamically from standard GOOGLE_CLOUD environment variables when not explicitly passed during LocalAgentConfig initialization.
Default Image Generation Model Optimization: Updated the default image generation model (DEFAULT_IMAGE_GENERATION_MODEL) to "gemini-3.1-flash-lite-image". Previous default image models often took too long to run on average during standard agent execution loops; this lightweight model ensures dependable, high-speed image generation by default while remaining fully replaceable via explicit model configuration if higher fidelity is required.
Local WebSocket Connections: Retries socket connections by resolving to both "localhost" and "127.0.0.1", ensuring dependable harness discovery under containerized setups.
Tool Runner Public Asyncness: Preserves original asynchronous and synchronous execution interfaces of tools when accessed via ToolRunner.get_public_callable.
Config Inheritance and Gaps: Cleaned up redundant base overrides in LocalAgentConfig and corrected Pydantic validation failures arising from None initialization variables.
Fixes(5)
Interactive Console Spinner Clobbering: Fixed a bug in run_interactive_loop where background spinner animation frames (⠼ Reasoning...) continuously clobbered user confirmation prompts (async_input and ASK_USER policy checks) every 80ms. The active spinner is now explicitly cleared (\r\033[K) and paused when an interactive input prompt opens, keeping confirmation lines readable and resuming the spinner smoothly after input is submitted.
Duplicate Tool Call Events: Fixed client rendering issues by filtering out custom tool events from StepUpdate payloads to prevent duplicate event dispatches.
SDK Idle Transitions: Corrected premature agent shutdown situations by properly checking and clearing idling flags once a TrajectoryStateUpdate signals transition to running.
LiteRT Connection Engine: Rectified local engine initialization bugs, including a missing protobuf import, a token constructor argument mismatch, and OpenAITool inheritance.
Shutdown Connection Handshake: Added extra execution buffer to local connections during agent shutdown, ensuring safe persistent state storage actions.
LiteRT model configurations, dynamic GCP environment hydration, and performance updates
This release expands local execution capabilities by broadening support for multimodal and web-connected tool workflows. Developers can now run Gemma models locally using LiteRT or integrate with local OpenAI-compatible APIs, natively return multimodal media from custom tools, and leverage a more robust, streamlined lifecycle hooks framework that is fully aligned with Model Context Protocol (MCP) workflows.
Improvements(4)
Local Model Connectivity: Introduced LiteRTAgentConfig and LiteRTConnectionStrategy for LiteRT-LM (supporting local Gemma execution), LocalOpenAIAgentConfig and LocalOpenAIConnectionStrategy for OpenAI-compatible APIs (supporting Ollama and LM Studio), and a background loopback HTTP translation server.
Multimodal Tool Outputs: Enabled custom tools to return media assets (Image, Document, Audio, Video) directly via a single tool response without needing separate follow-up turns (supplemental_media).
Built-in Web Fetch Tool: Integrated the read_url_content tool end-to-end for fetching structured web content natively with the ReadUrlContentResult Pydantic model.
MCP String Prefix Modernization: Decoupled tool calls and safety policy engines from legacy "mcp_" string synthesis, resolving name mismatch issues by utilizing explicit server_name attributes in tool evaluation.
Fixes(4)
Python 3.14 Compatibility: Resolved namespace class conflicts and typing normalization issues in agent.py and public_api_test.py under Python 3.14 deferred annotations evaluation.
Vertex Validation Errors: Cleared a misleading reference to API keys in VertexEndpoint validation error messages, limiting fields to project and location.
OTel Trace Warnings: Resolved detached contextvars warnings and set-status race conditions by removing use_span context managers from Turn/Session hooks and checking span recording readiness.
Exception Wrapping Mapping: Fixed agent_middleware integration check failures by making the example check for error message substrings instead of strict exception types.
Streaming tool calls, custom subagent delegates, and thread safety enhancements
This release introduces native OpenTelemetry tracing support, declarative subagent configurations, improved type safety, and critical robustness and compatibility updates.
Improvements(5)
OpenTelemetry Tracing Support: Integrates OpenTelemetry tracing into the SDK to translate session, turn, step, and tool lifecycle events into standard GenAI-compliant semantic spans for advanced monitoring and performance debugging, with custom task-safe active span propagation for tool execution.
Declarative Subagent Configurations: Added SubagentConfig and SubagentCapabilities in types.py to support constructing static subagents with declarative instructions and tools directly.
Type Safety in AgentConfig: Type-annotated policies, hooks, and triggers parameters on AgentConfig and its subclasses to improve type safety and overall developer experience.
Lifecycle Hook Routing: Shifted core orchestration of OnSessionStartHook, turn-level hooks (PRE_TURN and POST_TURN), and OnSessionEndHook to the connection layer, implementing the Python-side HookRouter for event routing.
Public API Cleanup: Hid internal validation methods on media and error classes by prefixing them with an underscore (_validate_mime_type and _from_pydantic on validation errors).
Fixes(2)
Historical Step Absorption: Ensured historical step absorption is properly drained during initialization to prevent persistence non-linearity issues in Conversation.
Python 3.14 Compatibility: Resolved potential name shadowing in the Conversation class by renaming the top-level connection module import.
Async tool execution buffers, memory state persistence, and error handling improvements
This release introduces major architectural refactorings, public API standardizations, and key new capabilities centered on centralizing model configurations to natively support multi-model backends, exposing a new built-in Google Web Search capability, enabling environment variable passing for Model Context Protocol (MCP) servers, and simplifying the agent initialization flow by removing dynamic runtime registrations.
Improvements(10)
Built-in Web Search Tool: Exposes the SEARCH_WEB tool directly within the SDK, enabling agents to leverage Google Search for grounded real-time information retrieval, complete with new developer examples (web_tools.py).
MCP Server Environment Variables: Added support for configuring and passing custom environment variables to launched stdio servers via the new env field in McpStdioServer.
Base URL and HTTP Headers Support: Out-of-the-box support for setting custom base URLs and HTTP headers.
Image Generation Aspect Ratio: Updated the SDK model config and wrapper to support specifying aspect_ratio within the image creation tool configuration.
Centralized Multi-Model Configuration: Replaces legacy singular gemini_config options with a unified, repeated models collection on AgentConfig and LocalAgentConfig to support multi-model routing, fallback strategies, and automated selection helpers.
Agent Session Lifecycle & API Standardization: Improves runtime safety by removing dynamic post-initialization hook and trigger registration in favor of session creation-time declarations.
Top-Level Package Exports: Exposed core SDK constructs (including Content, Image, Document, Audio, Video, from_file, BuiltinTools, and SystemInstructions) directly under the google.antigravity root module for easier access.
Hook Base Class Exports: Consolidated the base hooks implementation by exporting DecideHook, InspectHook, and TransformHook from the hooks package root.
Top-Level Policy Package: Created a new top-level policy package to clean up hook and workspace path validation dependencies.
Relocated Trigger Types: Moved the FileChange model and FileChangeKind enum from types.py to the specialized triggers package.
Initial subagent orchestration and standard Google AI model targets
This release introduces per-server MCP timeout configurations and improves local connection error handling.
Improvements(2)
Per-Server MCP Timeout: Added configuration support to set custom timeouts (in seconds) for individual MCP servers (BaseMcpServerConfig.timeout_seconds).
Terminal Error Propagation: The local connection now propagates terminal trajectory errors from the localharness binary as structured AntigravityExecutionError exceptions in the Python SDK during step collection.
This release adds Windows platform support, introduces programmatic turn-level stream cancellation, simplifies safety policy configurations for the Model Context Protocol (MCP), and removes the deprecated MCP SSE transport.
Improvements(3)
Windows Platform Support: Native compatibility added for Windows x86_64 and ARM64 environments. Path and file URI resolution now correctly handles drive letters and directory separators under Windows.
Programmatic Turn-Level Cancellation: Added programmatic stream cancellation via ChatResponse.cancel(). This programmatically aborts active generation turns directly from the client and raises AntigravityCancelledError (subclass of asyncio.CancelledError) to cleanly signal cancellation in the async flow (examples/getting_started/cancellation.py).
Direct MCP Safety Policy Configuration: Overloaded policy.allow, policy.deny, and policy.ask_user to accept server configurations (BaseMcpServerConfig) directly instead of typing namespaced string paths. Policy evaluation follows a 9-level precedence model (Specific > Prefix Wildcard > Global Wildcard) with longest-match prefix validation to protect against collisions.
Patches(1)
Deprecation of SSE Transport: Removed the legacy McpSseServer configuration and connection handlers in favor of standard Stdio and Streamable HTTP connection strategies.
This release focuses on significant enhancements to the Model Context Protocol (MCP) integration, adds native Vertex AI authentication support, improves robustness with better error handling, and includes critical fixes for token usage tracking.
Improvements(4)
MCP Tool Filtering & Simplified Policies: Added support for enabled_tools (allowlist) and disabled_tools (denylist) in server configurations, and overloaded safety policy helpers (policy.allow, policy.deny, policy.ask_user) to accept the MCP server configuration object directly.
Vertex AI Authentication: Integrated native support for Vertex AI authentication in the Python SDK.
MCP Tool Prefixing & Validation: The SDK now automatically namespaces and prefixes MCP tools (mcp_{server_name}_{tool_name}) to prevent name collisions when connecting multiple MCP servers. The name field is now mandatory in MCP server configurations and validated as a proper Python identifier.
Improved Error Handling: The SDK now raises explicit, descriptive exceptions for terminal errors rather than failing silently.
Fixes(2)
Structured Output Token Tracking: Fixed a bug where token usage_metadata was not correctly returned when structured output (response.structured_output()) was requested.
Type Checking Warnings: Fixed pytype warnings (including wrong-keyword-args in LocalAgentConfig) across several modules.
Queued Messages, File Attachments, and Performance Improvements
Support for queued messages, plain text, PDF, and markdown file attachments, model thinking level selector, and major conversation performance improvements.
Improvements(6)
Improved the ask questions panel: clearer option selection, and continue now only activates once you've answered the current question.
Added support for queued messages, including settings to configure execution behavior, a "Send Now" option, and an improved input card layout.
Added support for attaching and rendering plain text (.txt), pdf (.pdf) and markdown (.md) files in conversations.
Added a new built-in Antigravity Guide skill to provide helpful guidance when users ask about Antigravity.
Improved the model selector dropdown to choose thinking levels per-model (Low, Medium, High).
Updated icons for built-in slash commands and redesigned mentions/slash commands menu alignment in the chat input box.
Fixes(2)
Opening long conversations is now dramatically faster.
Reduced typing lag while the agent streams responses.
New model quota screen, agent permission and security fixes, and MCP stability improvements
Improvements(1)
Quota screen redesign: clearer, unambiguous view into “used” versus “remaining” credits in the Models tab of the Settings screen.
Fixes(4)
Fixed Agent Permission Issue: Fixed the bug where changes to the agent security mode were not persisted.
Improved MCP server stability: increased resilience to unresponsive MCP servers and improved browser agent self-troubleshooting abilities when using the Chrome DevTools MCP server.
MCP server schema compatibility: the mcp_config.json schema now accepts url in addition to serverUrl as a field.
New entries in the sensitive paths list: .vscode and .cache are now recognized as sensitive paths that require explicit user confirmation before the agent can access them in strict mode.
Fixed an issue where the Windows auto-updater fails to detect new releases. Please manually install the latest update if you are on version 1.16.5 or 1.18.3.
Adding the secure mode option, which enforces certain settings to prevent the agent from autonomously running targeted exploits and requires human review for all agent actions. Various security fixes.
With Nano Banana Pro, our agents have gotten even better at generating UI mockups, system diagrams, or relevant embeddable assets, all grounded in your existing codebase and knowledge.
Improvements(1)
Nano Banana Pro (incrementally rolling out)
Fixes(1)
Agent can now create scratch directories if no workspaces are open.
Patches(1)
Fixed an issue with the telemetry settings toggle on the settings page.
The original launch of Google Antigravity, with a fully-featured AI-powered IDE, new Agent Manager view, an integrated experience with Chrome, broad variety of rich Artifacts, user feedback flows, knowledge management, and much more. The vision for what development looks like in an agent-first paradigm.
