Quick Start
From install to your first agent turn in under a minute.
Install, sign in, and run your first turn in under a minute.
1. Launch in your project
The current working directory is the workspace root — agentty's filesystem tools won't touch anything outside it.
cd path/to/your/project
agentty2. Sign in
First launch opens an auth modal. Pick one:
- API key — paste an Anthropic
sk-ant-…token, or any provider's key. Saved to~/.config/agentty/credentials.json(mode0600). Unambiguously in-bounds, and works headless/over SSH. - OAuth (Claude Pro/Max) — opens your browser to sign in with your existing subscription; the callback writes the token to the same file.
✦TipNo Anthropic account? Run a local Ollama model with no key at all — press Esc, then Ctrl+P and pick ollama. See Authentication.
ℹNotePrefer a different model? Launch with --provider — e.g. agentty --provider openai -m gpt-4o or agentty --provider ollama for a local model. See Providers & Models.
3. Your first turn
Type a request, hit Enter. agentty streams the reply and lands tool calls inline. Mid-stream typing queues your next message and sends it when the current turn finishes. Esc cancels.
▌ add a --version flag that prints the build version and exits4. Pick a permission profile
You start in Write (autonomous) — every tool runs without a prompt, kept in-bounds by the sandbox and workspace boundary. Press S-Tab to cycle to Ask (writes, shell, and network prompt first) or Minimal (prompts for everything, reads included). Your choice persists.
Where to go next
- The Interface — what every part of the screen means.
- Keybindings — the full keymap.
- Providers & Models — Claude, GPT, Groq, or a local model.
- Tools — what agentty can actually do.
- SSH Air-gap — run on a box with no internet.