Models

• Anthropic
• Google
• OpenAI
• xAI
• Moonshot AI
• Qwen
• Z.ai

The model picker in the message box lists the models you can use. Each row shows the model's maker and whether it is Free or Pro, with its credits. Type in Search models... to filter the list.

We add models as they come out, so the picker always has the current list.

Click the model in the message box and choose one. The task runs on it, and we remember it as your preferred model for your next tasks.

Pro models need a paid plan. A model you cannot use shows a lock.

On paid plans, some models also have a Fast version. It is the same model, sent to the fastest provider that hosts it, so the agent answers sooner.

When a task starts, it uses the first of these that is set:

• The model picked in the message box for that task.
• Your preferred model: the last one you picked.
• The project's Model defaults for that kind of task.
• Our default for your plan.

So project defaults decide the model when nobody has picked one, for example for a teammate who has never opened the picker.

On paid plans, set Model defaults in the project's Settings: one model each for UI, API, Slack and Code tasks. Leave one on Platform default and we choose. See Projects.

A self-hosted instance can run on your own model instead of ours. Any OpenAI-compatible endpoint works, such as Azure OpenAI, a Bedrock gateway or vLLM. Set it with --llm-base-url, --llm-key and --llm-model when you install.