Configure AI

1. Open Settings → AI.
2. Pick a provider and paste your API key. For local inference, point Oarlock at your LM Studio, MLX, or Ollama endpoint.

Supported providers

• Bring your own key: OpenAI, Anthropic, Google (Gemini), Groq, xAI (Grok), Cerebras, OpenRouter, DeepSeek, Mistral, and any OpenAI-compatible endpoint.
• Local / offline: LM Studio, MLX, Ollama.

Where keys are stored

Keys go to the OS keychain: the macOS Keychain or the Windows Credential Manager. On Linux they go to a file in the app’s data folder that only your user can read. They never go to localStorage or the logs.

What the agent can do

• Plans, sub-agents, and custom agents with their own system prompt and tool subset.
• File read, write, edit, multi-edit, grep, and glob; bash with approval gating; background processes.
• A project’s AGENTS.md, skills (.claude/skills, .oarlock/skills) and .mcp.json servers are used once you allow them, and you’re asked again when they change.
• Local and remote MCP servers; each tool call is approved unless you trust the server.
• Spawn Claude Code in a terminal, inspect its output, and send follow-up work through approval-gated tools.
• In the composer, #handle inserts prompt snippets and @path attaches files. Voice input is built in.
• Plan mode generates a multi-step plan and confirms it before doing anything.