Short answer: install Gemini CLI, sign in with your Google or Cloud account, focus the terminal prompt, press ⌃+⌥+R (Mac) or Ctrl+Alt+R (Windows/Linux), speak your task, press again to insert, then send. AICHE does not run Gemini's tools or MCP servers.
Gemini CLI is Google's open-source (Apache 2.0) terminal agent: a reason-and-act loop with built-in tools plus local or remote Model Context Protocol (MCP) servers. It targets coding but also general terminal tasks (research, content, ops-style questions).
Important: Antigravity CLI transition (2026)
Google is unifying consumer and individual-tier agent tooling:
- June 18, 2026: Gemini CLI stops serving requests for Google AI Pro/Ultra, free Gemini Code Assist for individuals, and related IDE/GitHub individual flows. Those users move to Antigravity CLI (announcement).
- Unchanged: Organizations on Gemini Code Assist Standard/Enterprise, Google Cloud deployments, and paid Gemini / Gemini Enterprise API keys keep Gemini CLI access with ongoing model updates.
If you are on a personal Google account today, plan for Antigravity CLI and our /works-with/antigravity page. This Gemini CLI page stays accurate for enterprise and API-key workflows and for anyone still on Gemini CLI before the cutoff.
Why voice fits the CLI
Gemini CLI shares the same bottleneck as other terminal agents: short prompts get shallow tool use. The CLI exposes slash commands such as /memory, /stats, /tools, and /mcp (Gemini Code Assist agent mode docs) for power users, but the main win is still a dense natural-language first message: repo goal, files in scope, verification command, and non-goals.
Speaking that message is faster than typing it and tends to include more context (dependencies, rollback, "do not refactor X").
How it works
- Install per Gemini CLI documentation (package manager or guided install from Google developers site).
- Authenticate with the account type your org allows (personal license, Code Assist seat, or API key).
cdinto the project. Launch the CLI session.- Focus the prompt input in the terminal UI.
- Toggle ⌃+⌥+R / Ctrl+Alt+R, speak 30 to 90 seconds.
- Toggle again. AICHE inserts at the caret.
- Send. Gemini CLI chooses tools, MCP calls, and file edits per its policies.
Gemini Code Assist in VS Code: Agent mode in the editor is powered by Gemini CLI under the hood, but the input is the editor chat box, not the terminal. Same AICHE workflow: focus the VS Code Gemini chat field, hotkey, speak, insert. See /works-with/vscode.
Cloud Shell: Gemini CLI is available without extra setup in Google Cloud Shell; AICHE still runs on the machine where you installed the desktop app (typically your laptop inserting into a remote session only if focus and Smart Insert reach that field).
Honest limits
- Not Antigravity: Antigravity 2.0 desktop and Antigravity CLI are the successor product for many individual users. Features like Agent Skills, Hooks, and plugin migration are documented on Antigravity migration.
- Quotas: Gemini CLI and Gemini Code Assist agent mode share quotas on some editions (Google docs).
- Privacy: Enterprise data handling follows Code Assist Standard/Enterprise policies; individual tiers follow the individual privacy notice linked from the same doc set.
What you get with AICHE
- Custom vocabulary for GCP resource names, internal codenames, and package names.
- Auto-translation if you think in another language but want English prompts.
- Audio discarded immediately after processing, within 1 second.
FAQ
Q: Should I install Gemini CLI or Antigravity CLI?
A: If you are on a free or Google AI Pro/Ultra individual plan after June 18, 2026, Google points you to Antigravity CLI. Enterprise Code Assist users can stay on Gemini CLI.
Q: Does AICHE configure MCP?
A: No. You configure MCP in Gemini CLI; AICHE only fills text fields.
Q: Yolo / auto-approve modes?
A: Gemini CLI supports high-autonomy modes in some setups. Dictate explicit guardrails ("run tests but do not push") so auto modes have boundaries.
Related
Try it: one bug-fix brief with file path, repro command, and expected test output, inserted by voice before you send.