Use any model and any harness

PostHog Desktop lets you choose a harness and a model per task. The harness options are Pi, PostHog's own agent harness, plus Claude Code and Codex. Pick one in the composer before you send a prompt.

Pi

Choose Pi when you want PostHog's harness and the freedom to customize how the agent works. It can use:

• Skills from your machine, repository, and team
• MCP servers, including PostHog's server
• Specialized agents for focused research and planning
• Extensions that add tools, commands, and interface elements

Local Pi tasks use the custom resources on your machine. Cloud runs do not, so use a local task when an extension or local server is essential. See the Pi documentation to build an extension.

Other harnesses

Claude Code and Codex both run local tasks and cloud runs. Choose either harness in the composer. A cloud run can use a different harness than the task started with.

Choose a model

Open the model picker in the task composer before you send a prompt or while a task is running. The picker changes as providers add models, so it is the current source of truth.

PostHog Desktop offers models from Anthropic, OpenAI, and other providers. The model picker shows what is currently available.

See pricing for current rates and included usage.

Set reasoning effort

Models with extended thinking expose an effort level. Higher effort takes longer and can improve results on harder work.

The picker shows only the levels supported by the selected model.

The picker shows Fast mode when the selected model supports it. Fast mode reduces reasoning time for small, well-specified edits.

Authentication

PostHog Desktop authenticates each harness and model through your PostHog account. You do not need to manage API keys. Usage is billed to your organization.