TeamMemory turns your AI assistant into something that actually learns from past conversations instead of forgetting everything between sessions. It exposes seven MCP tools including memory_save for extracting structured knowledge from chat logs, memory_recall for semantic search across your team's experience database, and memory_context for pulling relevant history based on file locations. The system automatically scores experiences based on usage and lets them decay over time, so frequently referenced bug fixes and architecture decisions bubble up while outdated info fades. It runs on Postgres with Ollama embeddings for vector search, supports multi-user teams with RBAC, and includes a web UI for reviewing AI-extracted drafts before they get published. Useful when you want your AI pair programmer to remember that tricky API gotcha from three weeks ago.
claude mcp add --transport stdio ysydhc-team_memory -- uvx team_memory