Gemini CLI
by Google
Google's open-source terminal agent for Gemini — now enterprise-only after the Antigravity transition
Gemini CLI is Google's open-source, Apache-2.0 command-line AI agent that brings Gemini models into the terminal with file editing, shell execution, Google Search grounding and MCP tool support. Since 18 June 2026 it serves only Gemini Code Assist Standard and Enterprise licence holders and paid API-key users; consumer and free access moved to the closed-source Antigravity CLI.
Gemini CLI is an open-source AI coding agent that runs in the terminal, published by Google under Apache 2.0 and installed with npm install -g @google/gemini-cli or Homebrew. It wraps Google's Gemini models in a tool-using agent loop with built-in file operations, shell command execution, web fetching, Google Search grounding for fresh facts, conversation checkpointing and token caching, and multimodal input covering PDFs, images and sketches. Extensibility runs through the Model Context Protocol, so teams can attach internal services as tools alongside the community MCP servers, and authentication supports three distinct paths: personal Google OAuth, a Gemini API key, and Vertex AI for enterprises that need requests to stay inside their Google Cloud project. The repository has drawn roughly 106,000 GitHub stars with weekly preview, stable and nightly release trains and about 557 open issues. The commercial picture changed sharply in 2026. At Google I/O on 19 May 2026 Google announced it was consolidating developer tooling under the Antigravity brand, and on 18 June 2026 Gemini CLI stopped serving free-tier, Google AI Pro, Google AI Ultra and Gemini Code Assist IDE-extension users — breaking any script or CI job invoking the gemini command on those credentials. Enterprise access is explicitly unchanged for Gemini Code Assist Standard and Enterprise licence holders and Google Cloud customers on paid API keys, and the repository stays available under Apache 2.0 with security updates, but active development moved to the closed-source, Go-based Antigravity CLI. The announcement drew 297 downvotes against 6 upvotes from a community that had contributed to the open project, and Google acknowledged the successor would not reach day-one feature parity. Security has also been a live issue: an April 2026 advisory rated CVSS 10.0 described host command execution in headless CI before the sandbox initialised, fixed in 0.39.1.
Engineering organisations already standardised on Google Cloud with Gemini Code Assist Standard or Enterprise licences, who want a scriptable terminal agent that keeps inference inside their existing Vertex AI project and billing relationship.
A tool-using Gemini agent in the terminal and in CI, extensible over MCP and grounded in Google Search, without adding a new vendor to an estate that already buys Google Cloud.
At a Glance
- Category
- Developer Tools
- Pricing
- Subscription, Usage-based, Contact for pricing
- Target Market
- CTOs, Enterprise Developers, Platform Engineers, DevOps Engineers, Engineering Managers
- Deployment
- Open-source, API-based, Self-hosted, Cloud-first
- Founded
- 2025
- Headquarters
- Mountain View, United States
- Team Size
- 500+
Key Features
- ✓Terminal-native agent loop
Runs a reasoning-and-acting loop directly in the shell, reading files, editing code and executing commands without pulling the developer into a separate IDE or web UI.
- ✓MCP server support
Attaches internal APIs, databases and community Model Context Protocol servers as first-class agent tools, so the same CLI reaches company systems as well as the filesystem.
- ✓Google Search grounding
Queries live Google Search during a task so answers about libraries, APIs and errors reflect current documentation rather than only the model's training cut-off.
- ✓Multimodal input
Accepts PDFs, images and hand-drawn sketches alongside source files, which lets a developer turn a design mock or an architecture diagram into working code.
- ✓Three authentication paths
Supports personal Google OAuth, a Gemini API key, and Vertex AI for enterprises that need inference to stay inside their own Google Cloud project and billing account.
- ✓Checkpointing and token caching
Saves conversation state so long agent runs can be resumed, and caches tokens to cut the cost of repeatedly re-sending the same repository context.
- ✓Apache-2.0 open source
The full agent source is published under Apache 2.0 with weekly preview, stable and nightly release trains, so teams can audit, fork or pin a version.
Capabilities
Use Cases
- •Headless CI code tasks
Invoke the agent from a pipeline to apply a mechanical refactor, update dependencies or draft a fix, then open a pull request for human review.
- •Large codebase exploration
Ask the agent to trace a request path or explain an unfamiliar subsystem across many files before a developer commits to a change.
- •Internal tool orchestration over MCP
Wire company services into the agent as MCP tools so it can query a ticket system or an internal API during a task.
- •Grounded troubleshooting
Combine shell execution with live Google Search so the agent can reproduce an error and check current documentation in the same session.
- •Design-to-code from artefacts
Feed a PDF spec, screenshot or sketch to the agent and have it scaffold the corresponding component against the existing codebase conventions.
Ideal For
Best For
- ✓Google Cloud shops with Gemini Code Assist Standard or Enterprise licences that want terminal-native AI without a new procurement
- ✓Scripted and CI/CD automation where an agent must run headless against a repository rather than inside an IDE
- ✓Teams that need Vertex AI authentication so prompts and code stay inside their own Google Cloud project
- ✓Developers who want an Apache-2.0 agent they can read, fork and self-host against their own API key
- ✓Codebase exploration and multimodal tasks — reading PDFs, screenshots and diagrams alongside source files
Not Ideal For
- ✗Individual developers and hobbyists: the free Google-account path was switched off on 18 June 2026, and the replacement Antigravity CLI ships a far smaller daily allowance
- ✗Teams that want a long-lived, openly developed agent — active development moved to the closed-source Antigravity CLI and the Gemini CLI repository is now maintained mainly for enterprise users and security fixes
- ✗Organisations that need vendor-neutral model choice, since the agent is bound to Gemini models and Google's billing paths
- ✗Security-sensitive pipelines that cannot absorb agent-specific risk: a CVSS 10.0 remote code execution issue in headless CI was patched in 0.39.1, and researchers demonstrated sandbox escapes across this class of tool in July 2026
Integrations
Deployment
Market Analysis
Pros
- ✓Fully open source under Apache 2.0 with roughly 106,000 GitHub stars and weekly preview, stable and nightly release trains
- ✓Built-in Google Search grounding plus shell, file and web-fetch tools make it useful for real troubleshooting rather than code completion alone
- ✓MCP support means internal systems become agent tools without waiting on the vendor to build an integration
- ✓Vertex AI path keeps prompts and code inside the customer's Google Cloud project, which is often the difference between a pilot and an approved rollout
- ✓Enterprise access was explicitly preserved through the Antigravity transition, and the repository remains available with security updates
Cons
- ✗The free and consumer paths were switched off on 18 June 2026 with automation breaking on that date; scripts and CI jobs calling the gemini command on those credentials stopped working
- ✗Active development has moved to the closed-source Antigravity CLI, so the open repository is now effectively in maintenance for enterprise users and security fixes
- ✗Community reception to the transition was hostile — 297 downvotes against 6 upvotes on Google's own announcement, with contributors objecting to work they donated becoming enterprise-restricted
- ✗Google acknowledged the successor would not have full feature parity at launch, and users reported higher token consumption, lower quotas and missing features such as vim support and AI Studio authentication
- ✗Security exposure is real: an April 2026 advisory rated CVSS 10.0 allowed host command execution in headless CI before the sandbox initialised (fixed in 0.39.1), and July 2026 research demonstrated sandbox escapes across Cursor, Codex, Gemini CLI and Antigravity
- ✗Model lock-in — the agent is tied to Gemini and Google billing, with no first-class path to a non-Google model
Pricing
Open-source CLI with your own API key
$0 for the software
- ✓Apache-2.0 source
- ✓Bring your own Gemini API key
- ✓Metered per million input/output tokens
- ✓No free Google-account tier since 18 June 2026
Gemini Code Assist Standard
From $19/user/mo (annual)
- ✓Continued Gemini CLI access after the Antigravity transition
- ✓Reported ~1,500 requests/day
- ✓$22.80/user/mo on monthly billing
Gemini Code Assist Enterprise
From $45/user/mo (annual)
- ✓Continued Gemini CLI access
- ✓Reported ~2,000 requests/day
- ✓$54/user/mo on monthly billing
- ✓Admin controls and SLA
Vertex AI / Google Cloud paid API key
Usage-based
- ✓Inference stays in the customer's Google Cloud project
- ✓Billed per token by model
- ✓Enterprise access explicitly unaffected by the June 2026 change
The software is free and Apache-2.0; what you pay for is model access, and that changed materially in 2026. The generous free path — roughly 60 requests per minute and 1,000 per day on a personal Google account — stopped serving on 18 June 2026, along with Google AI Pro and Ultra. Continued access now requires either a paid Gemini API key or Vertex AI billed per million tokens, or a Gemini Code Assist licence; third-party reporting puts Standard at about $19 per user per month on an annual commitment ($22.80 monthly, ~1,500 requests/day) and Enterprise at about $45 annual ($54 monthly, ~2,000 requests/day). Google does not publish a Gemini CLI-specific price. Budget for reasoning tokens rather than list rates: agent sessions reload context repeatedly, so real spend tends to run well above a naive per-token estimate.
Security & Compliance
Connect
Sources
This page was written from 5 sources, 3 on domains other than github.com.
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
Raindrop
AI agent monitoring that catches silent production failures: Sentry for AI agents
GitLab Duo Agent Platform
Agentic AI across the whole GitLab DevSecOps lifecycle: planning, coding, code review, CI/CD and security agents under one governance model
CodeRabbit
AI code review and agentic change management for teams shipping human- and machine-written code
Qodo
Agentic code review and governance layer for teams shipping AI-generated code at scale