gnamiai — GnamiAI in your terminal

Talk to your hosted GnamiAI agent from PowerShell, bash, or zsh. Let it read your files, run shell commands, and edit code on your machine — all backed by your existing GnamiAI workspace (provider keys, skills, memory, integrations).

gnami login                                     # one-time, paste an API key
gnami                                           # interactive REPL
gnami "find the bug in src/api.ts and fix it"   # one-shot

Install

Windows (PowerShell)

iwr -useb https://gnamiai.live/install.ps1 | iex

macOS / Linux

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

Or via npm directly

npm install -g gnamiai

Requires Node.js 18+.

Sign in

gnami login

Opens gnamiai.live/app#/settings in your browser. Scroll to API keys, click New key, copy the gnami_… string (shown once), paste it back into the terminal. The key is stored in ~/.gnami/config.json (chmod 600 on Unix; user-only ACLs on Windows). Nothing else is persisted locally — no chat history, no caching, no telemetry.

To override at runtime (e.g. CI):

GNAMI_API_KEY=gnami_xxx gnami "your prompt here"

Use

Three modes:

gnami                      # interactive REPL — chat back-and-forth
gnami "your prompt"        # one-shot — print answer and exit
gnami chat "your prompt"   # explicit one-shot

In the REPL, slash commands work:

/help          show commands
/clear         wipe local history (server-side memory unchanged)
/cwd           show current working directory
/quit          leave (Ctrl-D also works)

What the agent can do on your machine

When the CLI is attached, the agent gets access to client-side tools that act on YOUR computer:

| Tool | What it does | Approval | |---|---|---| | read_file | Read a file's text content | Auto | | list_dir | List directory entries | Auto | | write_file | Create or overwrite a file | Prompt + diff | | edit_file | Find/replace inside a file | Prompt + diff | | bash | Run a shell command | Prompt |

By default, every write/edit/bash prompts you for approval (y/N). Destructive bash patterns (rm -rf, sudo, curl … | sh, mkfs, dd if=) always prompt, even with --yes.

Path safety: file ops are sandboxed to the current working directory by default. Pass --no-sandbox to allow paths outside cwd.

gnami --yes "add a TypeScript test for the login flow"
        ↑ auto-approve safe writes/edits/bash
gnami --no-sandbox "read /etc/hosts and tell me what you see"
        ↑ allow paths outside cwd

Web-side tools (web_search, fetch_url, gmail_draft, etc.) work as usual — they execute on the GnamiAI server, not on your machine.

Examples

Refactor a function

gnami "rename UserService.getById to UserService.findById everywhere"

Run tests + diagnose a failure

gnami --yes "run npm test and fix the first error you see"

Set up something new

gnami "scaffold a Vercel serverless function at api/hello.ts that returns the current time"

Read + summarize

gnami "what does src/repl.ts do? explain in 5 bullet points"

Environment variables

| Var | Default | Purpose | |---|---|---| | GNAMI_API_KEY | (saved key) | Override the saved Bearer token | | GNAMI_BASE_URL | https://gnamiai.live | Override the API endpoint | | NO_COLOR | (off) | Disable ANSI color output | | GNAMI_DEBUG | (off) | Print full stack traces on error |

Uninstall

gnami logout       # forget saved key
npm uninstall -g gnamiai

~/.gnami/config.json is removed by gnami logout. The directory itself stays so you can re-login without re-creating it.

Architecture (for the curious)

The CLI is a thin client:

  you ─type→ gnami REPL ─POST→ /api/agent/turn ─→ GnamiAI agent
                ↑                    │
                │                    │ emits client_execute action
                │                    ▼
                └─execute locally──── { read_file, write_file, bash, … }
                       (with approval gates)
                                     │
                                     ▼
                  POST follow-up turn with the tool result
                                     │
                                     ▼
                                next iteration

All your provider keys, skills, subagents, memory, and integrations live server-side in your GnamiAI workspace. The CLI doesn't replicate any of that — it's just another window into the same agent that runs the web dashboard.

License

MIT.