AIWiki
Malaysia
Back to all articles
Tools & Platformscoding agentscommand lineopen source

Gemini CLI

4 min readUpdated October 2026
Gemini CLI
Type
Open-source terminal AI agent
Developer
Google
Released
25 June 2025
Licence
Apache 2.0
Stack
Node.js, TypeScript, React (Ink)
Free tier
60 requests/min, 1,000 requests/day
Related
Claude Code, Codex CLI, Model Context Protocol
Gemini CLI is an open-source artificial intelligence agent that runs in the terminal, connecting Google's Gemini models to a developer's local files, shell and tools. Released by Google on 25 June 2025 under the Apache 2.0 licence, it handles coding, debugging, content generation, research and task automation through natural-language prompts, and doubles as a free front end to Gemini for anyone with a personal Google account.[1]

Background

Gemini CLI launched during the rapid emergence of terminal-first coding agents — tools that operate directly on a project directory, reading files, proposing edits and executing commands rather than merely chatting. Google positioned it as the "most direct path from your prompt to our model", open-sourced at github.com/google-gemini/gemini-cli so the community could inspect its behaviour and security.[1]

At launch it defaulted to Gemini 2.5 Pro with its 1-million-token context window and the largest free allowance in the category: 60 model requests per minute and 1,000 per day at no charge. Google integrated it with Gemini Code Assist, so the free licence also covers prompt-driven coding in VS Code.[1] Later releases added Gemini 3 Pro and Gemini 3 Flash — Google reported Gemini 3 Flash scoring 78% on SWE-bench Verified for agentic coding.[3]

Key Features

  • 1M-token context window. The agent can often hold an entire codebase in one session instead of working file-by-file.[3]
  • Reason-and-act loop over local tools. Gemini CLI reads and writes files, runs shell commands (confirmed by default), fetches URLs and grounds answers in live Google Search results.[1][4]
  • Model Context Protocol (MCP) support. Custom MCP servers (databases, Jira, design systems, internal APIs) are configured in ~/.gemini/settings.json and expose their tools to the model.[4]
  • Approval modes. Three tiers — default (confirm every edit and command), auto_edit (approve edits, confirm commands) and --yolo (approve everything) — plus a read-only plan mode.[3]
  • Extensibility. Built-in subagents (each running its own context loop), a TOML policy engine allowing or denying specific tools, optional OS-level sandboxing, session checkpointing, and headless mode for CI pipelines and scripts.[3][5]
  • Protocols. Beyond MCP, it supports Google's Agent-to-Agent (A2A) protocol for delegating to remote agents, and an Agent Client Protocol (ACP) mode that lets editors drive it over JSON-RPC.[5]
The project is built as an npm monorepo in TypeScript — npm install -g @google/gemini-cli on Node 20+ — with a weekly release cadence.[4]

Why It Matters for AI

Gemini CLI exemplifies the shift from chat interfaces to agentic workflows: the model acts on a real environment — files, shell, web — under human approval rather than only generating text. Its open-source licence and generous free tier made terminal agents accessible beyond paying subscribers, intensifying competition with Claude Code and OpenAI's Codex CLI. It also demonstrated that open protocols (MCP, A2A) have become the connective tissue of the agent ecosystem: Google adopted Anthropic's MCP rather than inventing a proprietary alternative. Security has grown with prominence — an April 2026 advisory rated a headless-mode prompt-injection flaw CVSS 10.0 before Google patched it in v0.39.1, a reminder that agents with shell access are a new attack surface.[3]

>Key Takeaways

  • Gemini CLI is Google's Apache-2.0 open-source terminal agent, released 25 June 2025.[1]
  • It combines a 1M-token context window, local file and shell tools, Google Search grounding, and MCP extensibility.[4]
  • Free personal accounts get 60 requests/minute and 1,000/day — one of the most generous free tiers among coding agents.[1]
  • Approval modes and a policy engine keep a human in the loop; headless mode extends it to scripts and CI (with documented injection risks).[3]
  • For Malaysia, it is a low-cost on-ramp to agentic development and, being open-source, a PDPA-compatible option for controlled enterprise use.

See Also

References

🇲🇾Malaysian Context

🇲🇾 Relevance to Malaysia: Gemini CLI's free 1,000-requests-per-day tier is directly relevant to Malaysian students, freelance developers and SMEs building AI capability without purchasing API credits — a real constraint in a market where paid tooling subscriptions are priced in US dollars. It lowers the barrier for the hands-on, project-based training that Malaysia's AI talent agenda and HRD Corp-claimed programmes emphasise.

For local engineering teams, the Apache 2.0 licence means the agent can be audited and self-hosted against enterprise endpoints — useful for PDPA-governed firms in banking, healthcare and the public sector that must keep source code and personal data within controlled environments. Google's regional data-centre presence and Malaysia's expanding cloud infrastructure also improve latency for terminal sessions run locally. Malaysia's large community of Python and JavaScript developers adopting agentic tooling fits the wider NAIO and MDEC push toward AI-assisted software development as a core national skill.

References

  1. ↑"Introducing Gemini CLI: your open-source AI agent." Google Blog, June 2025 - https://blog.google/innovation-and-ai/technology/developers-tools/introducing-gemini-cli-open-source-ai-agent/
  2. ↑GEMINI.md, google-gemini/gemini-cli repository (architecture and project context) - https://github.com/google-gemini/gemini-cli/blob/main/GEMINI.md
  3. ↑"Gemini CLI Review 2026." hokai.io - https://hokai.io/hub/agents/gemini-cli
  4. ↑"Gemini CLI: Complete Guide (2026)." Codersera - https://codersera.com/blog/gemini-cli-complete-guide-2026
  5. ↑"Gemini CLI: features, protocols, quickstart and alternatives." multiagentguide.top - https://multiagentguide.top/tools/gemini-cli
Back to all articles