Mike: Open-Source Legal AI with Self-Hosting, Multi-Model Support
Mike is an open-source legal AI platform positioned as a drop-in replacement for Harvey and Legora. It provides the same feature set without enterprise contracts or per-seat pricing; you only pay the model API costs. The codebase is fully auditable and forkable.
Core Features
- Assistant: Chat interface that reads uploaded documents, cites verbatim, runs multi-step workflows, and drafts/edits contracts end-to-end. Supports plugging in Claude or Gemini keys for full model control.
- Projects: Matter-scoped workspaces that maintain full context across conversations and documents (credit agreements, SPAs, leases, diligence packs).
- Tabular Review: Spreadsheet-style extraction across hundreds of documents in parallel. Every cell is cited back to a page and quote — no hallucinations.
- Workflows: Save proven prompts as reusable templates (CP checklists, credit agreement summaries, change-of-control reviews). Juniors can run them in one click firm-wide.
Why Open Source
- Zero license cost, no vendor lock-in, no rising per-seat fees.
- Self-hostable by design — documents stay within your infrastructure for compliance, data residency, and privilege control.
- Fork, extend with practice-specific workflows, integrate your DMS, or wire in your own citation engine.
- Audit every line: no black boxes around prompt construction, citation parsing, or data flow.
Start in the cloud or clone the repo. Available as a hosted product or to self-deploy from source.
📖 Read the full source: HN AI Agents
👀 See Also
RubyLLM: One Ruby Framework for All Major AI Providers
RubyLLM provides a single Ruby framework for OpenAI, Anthropic, Gemini, Ollama, and 800+ models. Features chat, vision, audio, tools, agents, streaming, and Rails integration.
P2PCLAW: A Peer-to-Peer Network for AI Agents to Publish Formally Verified Science
P2PCLAW is a peer-to-peer network where AI agents and human researchers can publish scientific results validated through formal mathematical proofs in Lean 4. The system uses GUN.js and IPFS, with post-quantum cryptography and privacy features for secure participation.
Need MCP Server Provides Semantic Tool Discovery for AI Agents
An MCP server called Need enables semantic search over 10,000+ tools from brew, npm, pip, and cargo. When an agent requests a task like 'compress these PNGs,' it finds pngquant, installs it, runs it, and reports back on success.
Claude's Silent Drop-Off: The Action Layer Failure When AI Agents Hit Business Sites
Claude can read business sites (pricing, booking flows, forms) but fails at the action layer — booking, submitting, or routing — due to lack of callable endpoints. This causes invisible user drop-off with no analytics signal.