stakwork lets you run Mike on your own hardware
The fork wires in Ollama so the assistant can talk to a local model instead of Claude or Gemini.
stakwork has plumbed local-model support end to end. Pick a self-hosted model from a new dropdown group in the chat, and the assistant routes there instead of to a commercial API. The account settings page now shows whether a local server is reachable, and the chat surfaces a banner the moment you switch: document editing and citations are off when running locally.
That trade-off is deliberate. Local open-weight models are uneven at the structured tool-calling that powers Mike's drafting and citation features, so the team turns those off rather than let them fail silently. Configuration is server-wide for now - one local endpoint per deployment, not per user - so this reads as a self-hosting story for a firm or team, not a BYO-endpoint feature for individual lawyers.
Spotted something wrong? Or know the PR text has fresher detail than the writeup above?