Skip to main content
Command line

Your AI agent in your terminal.

Install in 30 seconds. Login with your existing GnamiAI account. The agent reads your files, runs shell commands, and edits code on your machine — backed by the same provider keys, skills, memory, and integrations as your web dashboard.

   ▄████  ███▄    █  ▄▄▄       ███▄ ▄███▓ ██▓ ▄▄▄       ██▓
  ██▒ ▀█▒ ██ ▀█   █ ▒████▄    ▓██▒▀█▀ ██▒▓██▒▒████▄    ▓██▒
 ▒██░▄▄▄░▓██  ▀█ ██▒▒██  ▀█▄  ▓██    ▓██░▒██▒▒██  ▀█▄  ▒██▒
 ░▓█  ██▓▓██▒  ▐▌██▒░██▄▄▄▄██ ▒██    ▒██ ░██░░██▄▄▄▄██ ░██░
 ░▒▓███▀▒▒██░   ▓██░ ▓█   ▓██▒▒██▒   ░██▒░██░ ▓█   ▓██▒░██░
  agent in your terminal · v0.1.0
  signed in
  cwd: ~/projects/myrepo
  /help for commands · Ctrl-D to exit

 refactor src/api.ts to extract the auth logic into a separate file
⏳ thinking…
  I'll read the current file, extract the auth helpers, and create
  src/auth.ts with a clean export surface.

   Read src/api.ts (4,128 chars)
   Wrote src/auth.ts (1,902 chars)
   Edited src/api.ts
    skills: 2 · tool iter: 3 · llm calls: 4

 run the tests
⏳ thinking…
   bash ok
  All 12 tests pass.

Install

npm — recommended

npm install -g gnamiai

macOS / Linux

curl -fsSL gnamiai.live/install.sh | bash

Windows · PowerShell

iwr gnamiai.live/install.ps1 -OutFile install.ps1
.\install.ps1

Requires Node.js 18+. The wrapper scripts only run npm install -g gnamiai and verify the binary lands on your PATH — if you'd rather skip them entirely, the npm command above does the same job on every platform. Read install.ps1 before you run it.

Windows Defender flagging the script? That's a heuristic match, not malware — here's why and how to verify it.

Sign in & chat

  1. gnami login — opens the dashboard, mint an API key, paste it back.
  2. gnami — interactive REPL.
  3. gnami "your prompt here" — one-shot mode for scripting.

The key is stored in ~/.gnami/config.json with chmod 600. Override at runtime with GNAMI_API_KEY=… for CI pipelines.

What the agent can do on your machine

Five client-side tools the agent can call mid-turn. Read-only ops auto-run; anything that writes or runs prompts you for approval (with a diff preview).

read_file auto

Read a file's text content.

list_dir auto

List directory entries.

write_file y / N

Create or overwrite a file. Preview shown before writing.

edit_file y / N

Find/replace inside a file with a colored diff.

bash y / N

Run a shell command. rm -rf, sudo, and curl|sh always prompt.

File ops are sandboxed to the current working directory. Pass --no-sandbox to allow paths outside cwd. Pass --yes to auto-approve non-destructive operations.

Same agent, different surface

The CLI is a thin client. Everything that makes your GnamiAI workspace yours — the provider key you connected, the skills you wrote, the memory the agent has of past conversations, the Gmail / Calendar / GitHub / Slack integrations you authorized — lives server-side in your tenant. The CLI hits the same /api/agent/turn endpoint as the web dashboard.

That means you can mint an API key, drop into a terminal on a server you don't own, run gnami "deploy the staging branch", and get the same agent your web dashboard talks to. Switch back to the browser and the conversation is in your episode log.

Ready?

If you don't already have a GnamiAI workspace, sign up first — the CLI needs a Bearer key from your account.