Choosing a model

The agent's reasoning runs on a model you choose in Settings → Providers. API keys are stored in the macOS Keychain, never in plain config. Four provider types:

Claude Code CLI — no API key

If you already use the Claude Code CLI (claude) and are signed in, pick the "Claude Code CLI" provider and Laki Agent piggybacks on your existing Claude subscription — no API key to paste, nothing stored. Settings shows a live ✓ Detected badge with your CLI version once it finds it.

• The model runs via claude -p under the hood, using your logged-in session.
• It's fully sandboxed to a pure text turn — the CLI's own tools are disabled, and billing/auth environment variables are scrubbed so it always uses your subscription, never a stray API key.
• Leave Model blank to use your CLI's default, or pin one like sonnet.

Not detected? Install Claude Code, run claude once to sign in, then reopen Settings. The app resolves claude through your login shell, so it works even when launched from Finder.

Bring your own key

• Anthropic — paste an Anthropic API key and pick a Claude model.
• OpenAI — an OpenAI (or OpenAI-compatible) key + base URL.
• Laki LLM — point at your own self-hosted endpoint (e.g. a RunPod/vLLM /v1 URL) for a fully self-hosted model.

You can save several providers and switch the active one anytime; the agent uses whichever is active on its next run.