total-agent-memory

total-agent-memory is a local-first memory server for AI coding agents like Claude Code, Cursor, and any MCP client. It stores decisions, solutions, and project context so agents can recall them in future sessions without re-explanation. The system uses BM25 full-text search, local embeddings, fuzzy matching, and knowledge graphs to retrieve relevant information, with optional cross-encoder reranking.

Designed for developers using coding agents daily and researchers needing reproducible baselines, it runs entirely on your machine with zero LLM calls in the default recall path. The same package can also operate as a team server with personal, department, and company memory areas, complete with roles and audit trails. The maker says it was built with Codex CLI.