Released Jun 2025

Record sheet · No. 029
Gemini CLI
Google’s terminal agent for Gemini, Apache-2.0, with the unpaid tier handed to Antigravity CLI.
Project mark, taken from geminicli.com on 2026-10-01.

What it is
Google’s Apache-2.0 terminal agent for Gemini, launched on 2025-06-25 with a free tier of 60 model requests a minute and 1,000 a day. On 2026-06-18 the unpaid tier was replaced by Antigravity CLI; the repository stayed open and kept shipping nightly releases.
Who built it
Gemini CLI is Google’s own project rather than an individual’s: it sits in the google-gemini organisation, and the launch post was signed by two of the company’s staff — Taylor Mullen, a senior staff software engineer whom Google’s Cloud blog calls the creator of the CLI, and Ryan J. Salva, a senior director of product management for developer tools who came from GitHub and Microsoft. Salva has described how the team uses it: requirements written in Markdown first, 70% to 80% of his own work done in the terminal, and the team’s conventions kept as documents the model reads. It shares its technology with Gemini Code Assist, draws 699 outside contributors, and Google does not publish the team’s size; the unpaid tier was moved to Antigravity CLI on 18 June 2026.
Build log
5 stages- 01
A repository in April, a launch in June
The repository was created on 2025-04-17, and Google announced the project on 2025-06-25 in a blog post titled “Gemini CLI: your open-source AI agent”, authored by Taylor Mullen, a Senior Staff Software Engineer, and Ryan J. Salva, a Senior Director of Product Management. The npm package
@google/gemini-clipublished version 0.1.0 at 2025-06-25T13:03:28Z, with 0.1.1 and 0.1.2 following the same day. The Hacker News thread on 2025-06-25 drew 1,428 points and 788 comments. The earliest entry in GitHub Releases is taggedearly-accessand dated 2025-06-24, one day before the announcement. - 02
Free at launch, under Apache 2.0
The launch announcement offered 60 model requests per minute and 1,000 requests per day at no charge, and the README carries the same two numbers. Google’s stated reason for publishing under Apache 2.0 is that developers can inspect the code to understand how it works and verify its security implications, and the announcement describes Gemini CLI and Gemini Code Assist as sharing the same technology. The model named at launch was Gemini 2.5 Pro; the README now lists the Gemini 3 series. Releases run on three channels, all in UTC: preview on Tuesdays at 23:59, stable on Tuesdays at 20:00, and nightly every day at 00:00.
- 03
A file-deletion report, then official integrations
On 2025-07-22 a Hacker News post titled “I watched Gemini CLI hallucinate and delete my files” drew 304 points and 365 comments. Google announced GitHub Actions support on 2025-08-07 through the separate
run-gemini-cliAction, which reviews pull requests, triages issues and can be called with@gemini-cliinside an issue or a pull request. On 2025-09-24 the company announced that Google AI Pro and Ultra subscribers get Gemini CLI and Gemini Code Assist with a higher model limit. Extensions were opened for building on 2025-10-08, and an interactivity upgrade followed on 2025-10-15 and 2025-10-16. - 04
The unpaid tier is replaced by Antigravity CLI
A banner on the project’s documentation site reads: “Unpaid tier and Google One users: Gemini CLI was replaced by Antigravity CLI on June 18th, 2026.” Google had announced Antigravity 2.0 at I/O on 2026-05-19, and the project’s GitHub Discussions carry a thread from 2026-03-19 titled “Gemini CLI: mitigating abuse and prioritizing traffic”. The repository was not archived and stayed Apache-2.0: it still had a commit on 2026-09-25 and shipped a nightly version on 2026-09-26. The Plans page continues to list a “Free / For explorers” tier among six paths (Free, Google AI Pro, Google AI Ultra, Google Developer Program, AI Studio and Vertex AI) with the AI Studio tier annotated “Free tier data may be used to improve products”.
- 05
Where it stands
GitHub’s releases API returns 100 versions, which is that endpoint’s page limit; npm records 762 versions of the package, with 0.61.0 as the current stable version and 0.63.0-nightly.20260926.g2fe7c2d3f as the nightly. As of 2026-09-27 the GitHub API reports 107,167 stars, 14,634 forks, 812 open issues and 699 contributors, with the most recent commit dated 2026-09-25. The repository belongs to the
google-geminiorganisation and carries the topics ai, ai-agents, cli, gemini, gemini-api, mcp-client and mcp-server. The GitHub Actions integration lives in a separate repository,google-github-actions/run-gemini-cli.
Adjacent records
All records →No. 037
Headroom
A compression layer that sits between an agent and the model it calls: tool output, logs, file reads, RAG chunks and conversation history are shrunk on the machine that runs it, and the originals are kept in a local store so the model can ask for any block back.
No. 101
Phone Harness
A Python library and an agent skill that put a real iPhone or Android phone under a coding agent’s hands: the agent reads the screen through OCR or the accessibility tree, taps and types through HID events or adb, and then looks again to check what happened. Nothing is installed on the phone and there is no jailbreak — on a Mac the phone is an iPhone Mirroring window, elsewhere it is adb over USB or Wi-Fi.
No. 025
Codex CLI
OpenAI’s coding agent for the terminal, written in Rust after a rewrite from TypeScript.


