redleaves/context-keeper

# Technical Summary Implements a two-stage LLM-driven retrieval architecture—wide recall across semantic, timeline, and knowledge graph indices (~100 candidates) followed by LLM-based ranking—to surface contextually relevant information despite the recall-precision tradeoff. Provides unified four-dimensional context management (project, topic, conversation, code) with automated lifecycle tracking via MCP tools (session management, context retrieval, conversation storage, file association), enabling stateful interactions across conversation boundaries. Built in Go, MCP-compatible, and designed for developer teams to reuse architectural decisions and debug patterns without repeating project explanations to AI.

140 stars.