Built-in AI chat

1Using the chat

• Cmd+J shows or hides the chat; Cmd+T opens a new chat tab.
• Pick the model at the bottom of the chat.
• Attach images (PNG, JPEG, GIF, WebP, SVG) or a DESIGN.md file.
• Choose a style from the library, how many agents work in parallel, and whether the model thinks first (Auto, On or Off).
• Every tool call the agent makes is listed as a step. The stop button ends a run.

2Use your own API key

Open Settings (Cmd+,), Agents. Anthropic, OpenAI, OpenRouter, DeepSeek and Google Gemini are listed first. Also built in: MiniMax, Zhipu, GLM Coding Plan, Kimi (Moonshot), Bailian (DashScope) and Bailian Coding Plan, DouBao Seed, Ark Coding Plan, Xiaomi MiMo, ModelScope, StepFun and StepFun Coding Plan, and NVIDIA NIM.

Any other service works as a custom provider: give a base URL, a key and a model, in OpenAI-compatible or Anthropic format. Fetch models lists what the key can use.

3Use an agent you already have

If Claude Code, Codex CLI, OpenCode, GitHub Copilot, Antigravity, Grok Build or DeepSeek Harness is installed, the app finds it and the chat can run on it with your existing login. Connect or disconnect each one and hide models you do not use.

Agents that speak ACP work too. Kimi CLI, Gemini CLI and Qwen Code are preset; Add ACP agent takes any other command or URL.

4Photos and generated images

Settings, Images sets where pictures come from. Photo search uses Openverse, optionally with your own OAuth client. Image generation can use OpenAI, Google Gemini, Replicate, Atlas Cloud or a custom endpoint.

5Where keys are kept