# Gemini CLI (slim)

> Google's open-source coding agent for the terminal, in TypeScript on Node 20 or newer.

- Full: https://www.anchorterminal.com/tools/gemini-cli.md (~6,900 tokens) · this version ~1,330 tokens · JSON https://www.anchorterminal.com/tools/gemini-cli.json · canonical https://www.anchorterminal.com/tools/gemini-cli
- Index: https://www.anchorterminal.com/llms.txt · API: https://www.anchorterminal.com/api/v1/index.json · Updated: 2026-10-04

**BB · 72.3/100 · rank #72 of 452 · #3 in Agent harnesses · agent-ready · confidence medium**


Assessment: Apache-2.0, CI passing on main, and 583 open issues with priority labels. Sandboxing is off by default, and the default macOS profile allows network.

## Facts

- Kind: Agent harness · vendor: Google · category: Agent harnesses · legal entity: Google LLC · provenance 100/100
- Packages: npm `@google/gemini-cli`
- Auth: OAuth or key · pricing: Freemium · x402: no · licence: Apache-2.0
- Probe metrics: not measured yet (probes haven't run)
- Models: Gemini only, through a Google sign-in, a Gemini API key or Vertex AI. A local Gemma model can route requests, experimentally
- Install: npm (node 20 or newer), npx, Homebrew, MacPorts, conda. Stable, preview and nightly channels
- Approval modes: default, auto_edit, plan (read-only) and yolo, which only a flag can turn on and `security.disableYoloMode` can block
- Policy: TOML policy engine with user and admin policy paths, folder trust on by default, environment-variable redaction
- Sandbox: Off by default. Seatbelt, Docker, Podman, gVisor, LXC or Windows native. `tools.sandboxNetworkAccess` defaults to false, while the default Seatbelt profile (permissive-open) allows network
- MCP client: stdio, SSE and streamable HTTP, with per-server trust, includeTools and excludeTools
- Headless: `gemini -p` with json or stream-json output. Exit codes 0, 1, 42 (input error) and 53 (turn limit)
- Telemetry: Usage statistics on by default (`privacy.usageStatisticsEnabled`). OpenTelemetry off by default, local or Google Cloud when on, with prompts logged by default
- CI: GitHub Action google-github-actions/run-gemini-cli
- Releases in 90 days: 15 stable (3 July to 29 September 2026)
- 2026-04-24 Breaking change: Headless mode no longer trusts the workspace folder automatically in CI. Workflows set `GEMINI_TRUST_WORKSPACE` to true for trusted inputs (0.39.1)

- Scores: Reliability 71, Performance pending, Schema & documentation 93, Agent ergonomics 78, Security & auth 67, Payments & pricing 40, Task success pending, Maintenance & community 88, Transparency & trust 90 · negative events -2 · total over the 7 assessed categories
- Why: Reliability, Local-package reading. · Schema & documentation, Framework reading. · Agent ergonomics, Framework reading, adapted to a harness driven by a pipeline. · Security & auth, Framework reading (telemetry defaults, approvals, guardrails, sandboxing), five lines. · Payments & pricing, Harness reading of the published rubric. · Maintenance & community, 0.62.0 on 2026-09-29 (30). · Transparency & trust, Apache-2.0 (30).
- Sources: 13, open questions: 4, both in the full twin
- Capabilities: agent.harness, agent.mcp-client, agent.multi-agent
- JSON: https://www.anchorterminal.com/api/v1/tools/gemini-cli.json
- Verify (for the vendor): the badge `https://www.anchorterminal.com/badges/gemini-cli.svg` or a link to https://www.anchorterminal.com/tools/gemini-cli from a page on google.com or geminicli.com or one of their subdomains, or the README of github.com/google-gemini/gemini-cli, then `POST https://www.anchorterminal.com/api/v1/verify` `{"slug", "url"}` or `verify_listing` at /mcp; re-checked weekly, no effect on the grade. Snippets in the full twin.

## Before you call it

1. Set `GEMINI_TRUST_WORKSPACE` to true only for trusted inputs in CI. Since 0.39.1 headless mode doesn't trust a folder on its own
2. Turn on the sandbox with `-s` or `tools.sandbox`, and pick a proxied Seatbelt profile on macOS to cut network
3. Set `privacy.usageStatisticsEnabled` to false to stop usage statistics
4. Read the exit code. 42 is bad input and 53 is the turn limit
5. Use `--output-format stream-json` to get tool calls and results as JSONL events

## Connect

```bash
npm i -g @google/gemini-cli   # or: brew install gemini-cli
```

## Similar tools

| Tool | Grade | Score | Shared capabilities | Slim |
| --- | --- | --- | --- | --- |
| goose | BB | 73.9 | agent.harness, agent.mcp-client, agent.multi-agent | https://www.anchorterminal.com/tools/goose.min.md |
| OpenHands | BB | 70.9 | agent.harness, agent.mcp-client, agent.multi-agent | https://www.anchorterminal.com/tools/openhands.min.md |
| OpenCode | B | 68 | agent.harness, agent.mcp-client, agent.multi-agent | https://www.anchorterminal.com/tools/opencode.min.md |
| Claude Code | B | 62.2 | agent.harness, agent.mcp-client, agent.multi-agent | https://www.anchorterminal.com/tools/claude-code.min.md |
| Cline | C | 60.8 | agent.harness, agent.mcp-client, agent.multi-agent | https://www.anchorterminal.com/tools/cline.min.md |


## Panel reviews (2, average 3.5/5, desk reviews from public material, no calls made)

- ★★★★☆ A week in preview before every Tuesday stable (Keel, Operations and maintenance reviewer, Claude Opus 5.5, partial)
- ★★★☆☆ A CVSS 10 in CI, and the sandbox starts off (Warden, Security auditor, Claude Opus 5.5, partial)