MenteDB

MenteDB is an open-source, Apache 2.0-licensed Rust database built from the ground up as a cognitive memory engine for AI agents. Unlike vector database wrappers, it implements a purpose-built storage engine with HNSW indexing, graph-based belief propagation, and transformer-optimized context assembly. It ships a production MCP server with 32 tools across 6 categories, enabling Claude, Cursor, ChatGPT, and any MCP-compatible client to gain persistent memory with a single URL or CLI command.

• Sleeptime Enrichment — Background pipeline extracts semantic facts, links entities, detects communities, and builds user profiles from raw conversations automatically.
• Hybrid Search (BM25 + HNSW + RRF) — Combines keyword precision with semantic vector search for best-of-both-worlds memory retrieval.
• Contradiction Detection — Flags conflicting beliefs the moment they appear, preventing stale or inconsistent memories from polluting AI context.
• Pain Signals — Records negative feedback and past mistakes, surfacing anti-patterns before the AI repeats them.
• Bi-Temporal Validity — Memories carry valid_from/valid_until timestamps, enabling point-in-time queries and temporal invalidation instead of deletion.
• Phantom Memories — Detects knowledge gaps and creates placeholder memories so the AI knows what it does not know.
• Speculative Pre-Assembly — Predicts upcoming queries and pre-builds context windows, reducing latency and token waste.
• MCP Integration (32 tools) — Connect via stdio locally or Streamable HTTP remotely; setup takes one CLI command: npx mentedb-mcp@latest setup copilot.
• Native SDKs — Rust, Python, and TypeScript bindings available via cargo add, pip install, or npm install.
• Cross-Device Cloud Sync — Cloud mode keeps every device synchronized to the same memory store instantly.
• API Key Management & Webhooks — Dashboard-based key scoping, HMAC-signed webhook payloads, delivery logs, and automatic retries for production deployments.
• Token Efficiency — Compact format delivers 2.3x fewer tokens than raw JSON; delta savings of ~80% over 20 turns.