⚖MikeWatch Forks of the Mike legal-AI codebase, distilled for legal-tech readers
Home · punyaslokdutta/HarveyOss

punyaslokdutta/HarveyOss

A self-hosting-friendly take on Mike that trims cloud vendors and can run against a local model instead of a hosted API.

Two commits over a short burst in mid-May 2026, then quiet - a focused proof-of-concept for self-hosting that looks paused rather than actively developed.

View on GitHub →

This is punyaslokdutta's fork of Mike, aimed at people who want to run the thing themselves rather than wire it into a stack of hosted services. The work so far pulls in one consistent direction: fewer moving parts. Document storage gets folded onto the same platform that already handles login and the database, so there's one less outside account, set of API tokens, and bucket to manage.

The other half of the story is running Mike on your own terms. punyaslokdutta added a self-contained local setup that brings the frontend, backend, and storage up together, plus the ability to point Mike at a local, OpenAI-compatible model runtime like Ollama or vLLM. In practice that means you can stand up the whole stack - and even drive it with a model such as Qwen - without an OpenAI key.

If you're sizing up Mike for a private or on-prem deployment, this fork is worth a look. It's less about new legal features and more about making the plumbing cheaper and simpler to run. Click through to GitHub if you want the specifics.

What's in it

Direction

infrastructureintegration

Activity

Themed changes and pull requests touching this fork, newest first. Themed changes that haven't been turned into a public post yet still appear — they're real work even without a published writeup.

📝 Fork sync merge from origin/main (3 PRs + upstream willchen96 resync) 1 commit 2mo ago minor change

Threads of work (detailed view)

8 threads have been distilled into posts.

punyaslokdutta packages Mike to run on a laptop

A one-command local setup meant to make it easy to stand the whole app up on your own machine and kick the tyres before committing to anything hosted.

infrastructure
2 commits · latest 2mo ago

OpenAI-compatible local inference: Ollama and Qwen via env-configured base URL

punyaslokdutta makes Mike's OpenAI adapter point at any OpenAI-compatible endpoint - Ollama, vLLM, or the hosted service. Set `OPENAI_BASE_URL`, a model name, and an endpoint mode flag, and the same adapter talks to Ollama, vLLM, or any chat-completions-compatible runtime. A local runtime with no API key is treated as a valid provider, unblocking fully offline use.

infrastructuresecurity
2 commits · latest 2mo ago

Docker dev stack with Supabase and MinIO for zero-dependency local dev

punyaslokdutta adds a self-contained Docker Compose setup that brings up the full Mike stack locally - frontend, backend, Supabase auth/database, and S3-compatible blob storage - without any cloud accounts. The storage abstraction is also refactored so the same backend code can talk to Cloudflare R2 in production and MinIO locally.

infrastructureintegration
2 commits · latest 2mo ago

Pull requests (detailed view)

6 PRs touch this fork — inbound (filed against it) or outbound (filed from it). State icons match the editorial dashboard.

🟢 Open (2)

⛔ Closed without merge (4)