Overview / Description
HeurChain is an AI memory-infrastructure tool that gives AI agents persistent, structured memory across sessions, models, and machines, built for developers and teams running agents on the Model Context Protocol. Agents connect via MCP and gain a memory store organized in three tiers — ops/ (working memory), notes/ (session context), and self/ (long-term identity) — governed by an ACT-R cognitive-decay model, so memory activation fades with usage and stale information stops interfering with current context. Retrieval is hybrid, combining BM25 keyword search with Qdrant vector search via Reciprocal Rank Fusion, and the system reports sub-200ms P95 latency. HeurChain is MCP-native, integrating with Claude, Cursor, and Windsurf without prompt engineering, and supports session consolidation and working-group isolation. It can be self-hosted with Docker in under ten minutes or run as managed cloud. The project is open source under the MIT license. HeurChain is aimed at engineers who want agents that retain institutional knowledge and avoid re-explaining context every run.
Used For
Developers use HeurChain to give AI agents persistent, structured memory over MCP so they retain knowledge across sessions, models, and machines.
Pricing
Pros & Cons
Pros
- Three-tier memory (ops/notes/self) with ACT-R cognitive decay
- Hybrid BM25 + Qdrant vector retrieval via Reciprocal Rank Fusion
- MCP-native integration with Claude, Cursor, and Windsurf
- Sub-200ms P95 latency; Docker self-host in under 10 minutes
- Free Community tier; open source under MIT license
Cons
- Setup and self-hosting assume developer/Docker familiarity
- Token-based pricing on paid tiers can be hard to forecast
- Value depends on agents being run over MCP
- Overage fees apply once tier token limits are exceeded
Questions & Answers
Alternatives
Mem0, GPS, Zep, Letta, Pinecone