Gemini CLI Assess

coding-agentsdeveloper-aicligoogleagentsassessdevmay-2026gemini

Overview

Gemini CLI brings Google Gemini models to terminal workflows for scripting, refactoring, and automation (Gemini CLI).

It is an open-source (Apache 2.0) terminal agent with MCP support, project context through GEMINI.md files, checkpointing, optional sandboxing, headless mode for scripts and a GitHub Action for pull request review and issue triage. Authentication is via Google login, a Gemini API key or Vertex AI (Gemini CLI). On 19 May 2026 Google announced the transition of Gemini CLI to Antigravity CLI; since 18 June 2026 it only serves requests for Gemini Code Assist Standard and Enterprise licences and paid API keys (Google Developers Blog).

Assess for teams standardized on Google AI with the same credential and logging policies as other CLI agents.

Adoption Signals

  • Google's transition announcement keeps access unchanged for organisations using Gemini CLI via Gemini Code Assist Standard or Enterprise, Gemini Code Assist for GitHub through Google Cloud, or paid Gemini and Gemini Enterprise Agent Platform API keys (Google Developers Blog).
  • The repository states it remains available as an Apache 2.0 project "with no changes" (Transition discussion), and releases continue: v0.62.0 stable shipped on 29 September 2026 (Release v0.62.0).
  • Usage is large: about 107,000 GitHub stars and 14,700 forks (GitHub) and about 1.56 million npm downloads between 30 August and 28 September 2026 (npm download statistics).
  • An enterprise guide covers central system settings, coreTools allowlists, disabling YOLO mode, MCP server catalogues, enforced sandboxing, network proxy, telemetry and auditing, and restricting logins to corporate domains (Enterprise configuration).
  • Sandboxing options cover macOS Seatbelt, Docker or Podman, a Windows native sandbox and gVisor on Linux (Sandboxing).

Risks

  • Product transition. For unpaid and Google One users, Gemini CLI "was replaced by Antigravity CLI on June 18th, 2026" (Terms and privacy); skills, hooks, subagents and extensions carry over to Antigravity, but investment in Gemini CLI-specific configuration may need to be migrated later (Google Developers Blog).
  • Prompt injection through repository context. Tracebit showed that instructions hidden in a README could make the early release silently run and exfiltrate environment variables; Google fixed it in v0.1.14 on 25 July 2025 (Tracebit). Running the agent on untrusted code without a sandbox remains the main exposure.
  • Permissive local defaults. Sandboxing is opt-in (Sandboxing), and any user can switch on YOLO mode, which auto-approves tool calls, unless an administrator disables it through system settings (Enterprise configuration).
  • Data terms depend on the login path. Google account login, unpaid Gemini API keys, paid API keys and Vertex AI each fall under different terms of service and privacy notices (Terms and privacy); a developer who signs in with a personal account or an unpaid key can put company code under individual or unpaid-service terms, so personal logins should be blocked for company repositories.

Pros & Cons

Advantages

  • Apache-2.0 open source and actively released (v0.62.0 stable on 29 September 2026, with nightly and preview channels), with about 107,000 GitHub stars and roughly 1.5 million npm downloads in the month to 28 September 2026.
  • A documented enterprise configuration model: a system-level settings file, tool allowlists, the option to disable YOLO mode, enforced sandboxing, proxy control, telemetry and restriction of logins to corporate domains.
  • Remains supported for organisations on Gemini Code Assist Standard or Enterprise licences and for paid Gemini API or Gemini Enterprise Agent Platform keys, and includes MCP support, GEMINI.md context files and a GitHub Action.

Disadvantages

  • Google moved individual users to Antigravity CLI: since 18 June 2026 Gemini CLI no longer serves requests for free, Google AI Pro and Ultra users, so the product's future is tied to enterprise licences rather than broad community use.
  • An early flaw let prompt injection in untrusted repository files trigger silent execution of shell commands; it was fixed in v0.1.14 (July 2025), but the attack class remains relevant for any agent that reads repository context.
  • Sandboxing must be enabled explicitly, and the tool is designed around Gemini models, so model choice and data processing terms are bound to Google.

Recommendation

Assess Gemini CLI only if the organisation already holds Gemini Code Assist Standard or Enterprise licences or runs Gemini through Vertex AI, and wants a terminal agent inside that contract. For everyone else, the individual path now leads to Antigravity CLI, which should be assessed together with the Google Antigravity entry rather than starting new work on Gemini CLI.

For an assessment, deploy the enterprise system settings file first: restrict logins to the corporate domain, disable YOLO mode, define a coreTools allowlist and an approved MCP server catalogue, enforce a container or gVisor sandbox, and route traffic through the corporate proxy with telemetry sent to your own collector. Compare results against the terminal agents already on the radar, such as Claude Code, on the same tasks before deciding whether to move it forward.

Sources

Consulting

From radar to implementation

You have assessed the technologies — we help you bring them into production systems safely and measurably.

AI Trainings

AI Trainings

We make your teams AI-ready. Our trainings turn abstract buzzwords into applicable skills. Clear. Practical. Built for doers.

Learn more