# Antigravity CLI — Google's terminal agent: a keyboard-driven TUI that edits code, spawns subagents and gates every tool call. > Source: The Agents Index — https://theagentsindex.com/antigravity-cli (structured, researched, re-verified) > Facts last verified: 2026-09-10 Antigravity CLI is Google's terminal user interface for the Antigravity agent harness, installed as the agy binary on macOS, Linux and Windows. It runs multi-file agent tasks in the shell, spawning parallel subagents and pausing at artifacts for review. | Fact | Value | | --- | --- | | Website | https://antigravity.google/product/antigravity-cli | | Price | from €21.99/month | | Free tier | Yes | The 'For Individuals' plan costs $0/month with no subscription: the full agent model list, unlimited Tab completions and Command requests, and 'Basic weekly rate limits' whose numeric ceiling the vendor does not publish. ## What it does **Writes code** · Yes — The TUI's core loop is instructing agents in plain language to edit and build code across the workspace, with the same harness as the desktop app. > Edit, orchestrate, and build all in natural language. Tell your agents what you need, and they’ll work on getting it done. > — https://antigravity.google/product/antigravity-cli · Explore the main features, 'Work in Natural Language' section **Debugs and fixes** · Yes — A dedicated /boost command escalates hard defects to multi-agent reasoning, and the docs push you to give the agent a test command so it iterates on failures itself. > Multi-agent deep reasoning for complex bugs, race conditions, and algorithms. > — https://antigravity.google/docs/cli/features · Core Slash Commands table, /boost row **Writes tests** · Yes — The vendor's own best-practice sequence is to have the agent author a test block before implementing, then run the suite to verify its change. > If tests do not exist, direct the agent to write a standard test block first. > — https://antigravity.google/docs/cli/best-practices · Establish verification loops, numbered step 2 **Reviews code** · Partly — The diff viewer opens workspace changes, agent turns and any commit in the repository for line-level comments, but review is a human reading diffs to steer the agent; no pull-request or reviewer workflow is documented. > The /diff command opens the Interactive Diff Viewer, a full-screen panel that allows you to inspect changes in your workspace and conversation history. > — https://antigravity.google/docs/cli/commands/diff · Overview paragraph **Whole-codebase work** · Yes — It indexes the trusted workspace and the docs direct large refactors and repository-wide sweeps to concurrent background subagents rather than one serial thread. > For large-scale sweeps or multi-file refactoring, direct the primary agent to spawn concurrent background subagents. > — https://antigravity.google/docs/cli/best-practices · Automate and script, 'Fan out using parallel subagents' **Ships it** · Partly — The agent executes local shell commands under your permission rules, so it can run deploy scripts you authorise, but no release, CD or environment-promotion feature is documented anywhere in the CLI docs. > The Terminal Sandbox is a lightweight security isolation mechanism that protects your host system from potentially destructive file manipulations or unauthorized outbound network requests when the agent executes local shell commands. > — https://antigravity.google/docs/cli/features · Terminal Sandbox section, opening paragraph **Done-for-you service** · No — Self-installed software: you run the agy binary on your own machine. Google sells no build-and-run service around it. > Antigravity CLI runs natively on macOS, Linux, and Windows. > — https://antigravity.google/docs/cli/install · Installation & auth, Installation ## How much it does unattended **Runs autonomously** · Yes — Autonomy is a setting: the default prompts before writes and commands, while always-proceed or a headless one-shot run finishes a task with nobody watching. > Select agent autonomy level (request-review, always-proceed, or strict). > — https://antigravity.google/docs/cli/features · Core Slash Commands table, /permissions row **Multi-agent** · Yes — The main agent spawns concurrent subagents with their own tool grants, monitored in an /agents panel; collaborative agent teams via /teamwork-preview are gated to paid plans. > Subagents are independent, concurrent agent sessions designed to tackle specific background tasks in parallel with the main conversation. > — https://antigravity.google/docs/cli/features · Subagents in Antigravity CLI, 'What are Subagents?' **Agent permissions** · Yes — Every sensitive action is an action(target) resource sorted into deny, ask and allow lists with deny winning, and unconfigured commands, MCP tools and out-of-workspace files default to asking. > Every sensitive operation the agent performs is represented as a permission resource formatted as action(target). > — https://antigravity.google/docs/cli/permissions · Fine-grained permissions, opening paragraph ## Which models it runs on **Claude** · Partly — Claude Sonnet 4.6 and Opus 4.6 are selectable reasoning models on the free and consumer paid plans, but the model matrix marks both unavailable on Enterprise deployments. > | Claude Opus 4.6 (thinking) | ✅ | ✅ | ✅ | ❌ | > — https://antigravity.google/docs/models · Reasoning Model table, Claude Opus 4.6 row (Enterprise column) **GPT** · Partly — The only OpenAI-lineage option in the reasoning-model table is the open-weight GPT-OSS-120b, absent on Enterprise; no proprietary GPT model appears anywhere in that per-plan matrix. > | GPT-OSS-120b | ✅ | ✅ | ✅ | ❌ | > — https://antigravity.google/docs/models · Reasoning Model table, GPT-OSS-120b row (Enterprise column) **Open models** · Partly — One open-weight model, GPT-OSS-120b, is offered as a hosted choice on the consumer plans and is switched off for Enterprise; you cannot supply your own weights. > For the core reasoning model, Antigravity offers leading frontier models. Availability depends on your plan: > — https://antigravity.google/docs/models · Reasoning Model section, sentence introducing the per-plan model table **Model choice** · Partly — You pick the reasoning model from the vendor's fixed per-plan list and it persists across sessions, but auxiliary models are fixed and there is no third-party provider selection. > Users can select which reasoning model they want to use within the model selector drop-down under the conversation prompt box: > — https://antigravity.google/docs/models · Reasoning Model section, paragraph under the plan table ## Where you use it **In your editor** · Partly — The CLI itself is a keyboard TUI with no editor surface; the same agent harness runs in Antigravity 2.0 and IDE extensions, and conversations export between them. > | Primary interface | Keyboard-driven TUI | Visual desktop editor / IDE | > — https://antigravity.google/docs/cli/overview · Platform comparison table, Primary interface row **On the command line** · Yes — The terminal is the whole product: a lightweight TUI installed as agy that carries the desktop agent's reasoning and multi-file editing into the shell. > The Antigravity CLI is the lightweight Terminal User Interface (TUI) surface of Antigravity. > — https://antigravity.google/docs/cli/overview · Antigravity CLI Overview, first sentence **In your pipeline** · Yes — Headless print mode is built for pipelines, returning JSON or NDJSON events; an unauthenticated CI run fails fast rather than hanging, and a Gemini API key avoids the browser sign-in. > Run Antigravity CLI non-interactively to script agent tasks, integrate with CI pipelines, and capture machine-readable output. > — https://antigravity.google/docs/cli/headless · Headless mode, opening paragraph **In a browser** · Yes — The CLI ships a Remote Control daemon you register as an OS service, after which the browser dashboard drives that machine's conversations, tasks and artifacts. > The headless daemon is built into the Antigravity CLI. > — https://antigravity.google/docs/remote-control · Install Remote Control headless daemon, first line ## Whose machine it runs on **Self-hosted** · Partly — The binary and its sandbox run on your own machine and requests can be pointed at a Gemini-compatible endpoint or your own Cloud project behind VPC-SC, but the models themselves stay Google-hosted. > To send model requests to a different Gemini-compatible endpoint, set the GOOGLE_GEMINI_BASE_URL environment variable: > — https://antigravity.google/docs/cli/install · Using a Gemini API key, 'Point the CLI to a custom endpoint' **Open source** · No — Shipped as a prebuilt executable fetched by an install script; no source repository for the CLI is published on the site. > Execute the native installer script to download and install the executable to ~/.local/bin/agy > — https://antigravity.google/docs/cli/install · Installation, macOS and Linux ## What it costs to run **How it meters** · Yes — Billing is subscription quota first, then metered AI credits at Agent Platform consumption rates, or pure consumption billing for organisations; nothing is priced per seat in the CLI itself. > AI credits are consumed at standard Gemini Enterprise Agent Platform consumption pricing. > — https://antigravity.google/docs/plans · Overages section, first paragraph **Free tier** · Yes — A standing zero-cost individual plan needs no subscription and no sales call, giving the full agent model list with basic weekly rate limits rather than a time-boxed trial. > For Individuals $0/month > — https://antigravity.google/pricing · Plan cards, 'For Individuals' card heading and price **API access** · Yes — Headless runs return a JSON envelope or streaming NDJSON with token usage and tool calls, accept a JSON schema, and take prompts on stdin; note the terms forbid third-party clients using Antigravity OAuth. > Headless mode (also called print mode) sends a single prompt to the agent, streams or returns the response, and exits. > — https://antigravity.google/docs/cli/headless · Headless mode, second paragraph **MCP server** · Partly — It is an MCP client: local stdio and remote servers are configured globally or per workspace with per-tool approval, but no MCP server is documented that would let another agent query this CLI. > Antigravity CLI supports both local stdio processes and remote host MCP server configurations. > — https://antigravity.google/docs/cli/mcp · Antigravity CLI section, opening paragraph **Bring your own key** · Yes — You can run the CLI on your own Gemini API key with no account session, or on your own Google Cloud project; the plans page still says bring-your-own-key is unsupported, which contradicts this. > Run Antigravity CLI with your own Gemini API key instead of a signed-in Google account. Model requests go directly to the Gemini API, and the CLI never establishes an account session. > — https://antigravity.google/docs/cli/install · Using a Gemini API key, opening paragraph ## Buying it for a team **A company can buy it** · Yes — Organisations buy through Google Cloud rather than this site, connecting the CLI to a billed Cloud project or a Gemini Enterprise licence under Cloud terms. > Google Cloud customers can now access Antigravity 2.0 and CLI. > — https://antigravity.google/pricing · Plan cards, 'Organization plan via Google Cloud' card **Seat model** · Partly — Enterprise access is a per-user licence a Cloud administrator assigns, with one licence per project and location, but no seat price, seat minimum or ceiling is published anywhere on the site. > Licenses are assigned by your organization’s Google Cloud administrator. > — https://antigravity.google/docs/enterprise · Troubleshooting & Diagnostics, 'No Licenses Appear During Setup' **Pooled budget** · Partly — Consumer plans meter into your own credit pool rather than a team balance; organisations get consumption billed centrally to the Cloud project behind the licence, with no documented per-seat budget cap. > Gemini Enterprise Agent Platform - Connect directly to Agent Platform API to use Antigravity with consumption-based billing. > — https://antigravity.google/docs/enterprise · Overview & Key Benefits, two-ways list **Admin controls** · Partly — Central administrative control arrives with the Google Cloud path; on individual plans the allow/deny/ask rules live in a user-editable settings file, and the vendor says admin URL allowlists are not yet honored. > Operates under your existing Google Cloud Terms of Service with centralized administrative controls. > — https://antigravity.google/docs/enterprise · Overview & Key Benefits, 'Enterprise Governance' panel **Audit log** · Partly — Enterprise instances can log model requests and responses into the Cloud project for compliance records; nothing describes an administrative action trail, and consumer plans get no audit log at all. > Audit model interactions and maintain enterprise compliance records for your Gemini Enterprise Agent Platform instance. > — https://antigravity.google/docs/enterprise · Security & Governance, 'Request & Response Logging' **Single sign-on** · Partly — Business sign-in is a Google SSO flow that detects your licence, and external identity providers such as Okta work through Workforce Identity Federation; a documented defect logs some federated users out on restart. > Bring Your Own Identity (BYOID) uses Workforce Identity Federation (WIF) to let your organization authenticate through an external identity provider, such as Okta, instead of a standard Google Account. > — https://antigravity.google/docs/enterprise · Bring Your Own Identity (BYOID / WIF) section ## What happens to your code **Opt out of training** · Yes — Consumer terms say your interactions feed Google's model development unless you change a setting, and the enterprise path states corporate data is never used for training. > If you don’t want your Interactions used in this way, navigate to settings to change your preference on how such data is used. > — https://antigravity.google/terms · Additional Terms of Service, clause 5 **Data residency** · Partly — Enterprise deployments choose a global, US or EU endpoint to meet residency rules, with image generation only on global; consumer plans publish no jurisdiction choice at all. > Antigravity CLI, Antigravity 2.0, and IDE Extensions support multi-region deployment endpoints to satisfy regional data residency requirements: > — https://antigravity.google/docs/enterprise · Regional Endpoints & Capability Matrix, sentence above the table **Getting out** · Partly — Conversations live locally and export to the desktop app, but deleting your stored interactions means emailing support rather than a self-serve control, and nothing says whether purchased credits expire. > You will have the option to delete your Interactions. If you would like to request that your Interactions be deleted, you can email antigravity-support@google.com. > — https://antigravity.google/terms · Additional Terms of Service, clause 3 _Not established: Certifications_ ## Platform record (as of 2026-08-30) Reported by the code-hosting platform, not by the vendor. | Fact | Value | | --- | --- | | Stars | 2081 | | Forks | 189 | | Last push | 2026-08-27 | ## Provenance 1 of 23 sources behind the facts above are independent of the vendor. ## Coverage elsewhere Pages about this product that the vendor does not control. The label states how independent each one is. | Source | Kind | Read | | --- | --- | --- | | developers.googleblog.com | Independent source | 2026-08-23 | | github.com | Platform record | 2026-08-23 | | theregister.com | Independent source | 2026-08-23 | ## What changed - 2026-09-10 — Capabilities · Commercial terms · Free tier · Price - 2026-09-02 — Capabilities · Commercial terms · Free tier · Price - 2026-09-01 — Capabilities · Commercial terms · Free tier · Price - 2026-08-31 — Capabilities · Commercial terms · Free tier - 2026-08-23 — Alternatives · Documentation link · One-liner · Summary - 2026-08-19 — Summary ## Correcting this record Anything marked "Not established" is a question we have not answered — it is not a No. If you represent Antigravity CLI, or anything here is wrong or out of date, you can supply it with a source: the operations this site accepts from an agent are listed at https://theagentsindex.com/.well-known/agents.json — that catalogue needs no key to read, and each entry states its own auth. Supplying evidence is free. Nothing that can be bought here changes what a field says, whether a gap is closed, or how this record is scored.