Google Antigravity

• 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.
• 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

• 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.
• 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.

• Added a faster return path for enterprise sign-in, so a returning user is offered a single Continue with ... option naming the method they last used.
• Added support for OAuth client ID metadata documents when connecting to an MCP server, so servers implementing that part of the specification no longer require a manually supplied client ID or dynamic client registration.
• Improved the /context panel, which is now scrollable and sized to the terminal. /usage is sized the same way.
• Improved markdown-defined agents with a single inheritCustomizations switch that decides whether the agent adopts your skills, rules, plugins, subagents and MCP servers, replacing per-kind defaults that silently disagreed with each other.
• Improved the setting that allows access outside your workspace, which now grants only read access; writes are auto approved according to the cycle mode setting.
• Fixed the CLI exiting silently when its language server failed, both at startup and while running; the failure is now printed to the terminal and the process exits non-zero instead of reporting success.

0.1.13 August 18, 2026 Pre-tool argument modification, synchronous lifecycle hooks, RunCommandConfig timeouts, and step correlation IDs 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 ,

• Added support for GEMINI_API_KEY, so the CLI can run against the Gemini API directly without signing in. Set modelProvider: "gemini" in settings.json, export GEMINI_API_KEY, and point GOOGLE_GEMINI_BASE_URL at a custom endpoint if you need one. The banner and /help show Gemini API key as the credential, and /logout explains that it comes from the environment rather than appearing to end a session.
• Improved /codesearch resilience by falling back to a local search when the ripgrep binary cannot be executed, such as when an endpoint security agent blocks it.
• Improved artifact list scrolling by caching each file's preview line count, which previously re-read 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, so a custom agent can list and kill its own background tasks, and reworded its step lines to read Killed task X and Checked task X.
• Improved search_web steps by showing the query while the search is still running rather than only after it returns.

2.5.5 August 13, 2026 Windows Media Attachments and Chat Responsiveness Improvements Bug fixes addressing media attachment failures on Windows and improving message responsiveness when interacting with the agent. Improvements (0) Fixes (2) Fixed an issue where the agent failed when attaching images or media files on Windows machines. Improved responsiveness and reduced hanging issues when sending messages to the agent. Patches (0) 2.5.2 August 12, 2026 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.

0.1.12 August 13, 2026 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. Improvements (0) 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. Patches (0) 0.1.11 August 11, 2026 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

2.5.2 August 12, 2026 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. Patches (0) 2.1.1 June 22, 2026 Model quota screen and agent security fixes New model quota screen, agent

• 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.

0.1.11 August 11, 2026 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

• 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.

0.1.10 August 4, 2026 Prioritized inference tiers, real-time token usage streaming, and context-aware hook decorators 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

• 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.

• 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.
• 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.

• 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.