markhuangai/dense-mem

Description and use

**Brief Summary**
*markhuangai/dense‑mem* is a self‑hosted AI‑agent memory server written in Go that combines multi‑claim provenance (MCP), typed claims, conflict detection, embeddings, and recall with both PostgreSQL and Neo4j back‑ends. It lets developers turn ad‑hoc prompts and tool calls into repeatable, searchable agent workflows, making multi‑agent coordination and tool‑use pipelines far more reliable.

**Value**

• **Unified Memory Layer** – Stores structured “claims” with provenance, enabling agents to retrieve exact evidence and resolve contradictory information automatically.
• **Hybrid Persistence** – Uses PostgreSQL for fast key‑value storage and Neo4j for graph‑based reasoning, giving you the best of relational and graph queries in one service.
• **Embeddings & Recall** – Built‑in vector support lets agents perform semantic search over past interactions, dramatically improving retrieval‑augmented generation (RAG) performance.
• **Workflow Repeatability** – By exposing a clean API/SDK/CLI, the server makes it trivial to script multi‑agent pipelines, enforce consistent memory policies, and audit decision paths.

**Practical Adoption Path**

1. **Prototype** – Spin up the Docker image (or compile from source) and connect a simple Go or Python client to store a few typed claims. Verify retrieval and conflict‑detection behavior with synthetic prompts.
2. **Integrate** – Replace ad‑hoc in‑memory caches in existing agents with calls to the dense‑mem API; map your current PostgreSQL tables or Neo4j schema to the server’s claim model.
3. **Extend** – Add custom claim types or embed your own vector model via the provided SDK; hook the server into your orchestration layer (e.g., LangChain, CrewAI) to automatically persist tool outputs.
4. **Productionize** – Deploy the service behind a load balancer, enable TLS, and configure backup/replication for both PostgreSQL and Neo4j. Use the built‑in health endpoints and metrics to monitor latency and storage growth.

**Production Readiness**

• **Activity & Adoption** – Recent commits (as of 2026‑06‑22), 29 stars, 4 forks, and a Go codebase with 17 topical tags indicate an active community.
• **Maturity** – Core features (MCP, conflict detection, embeddings, dual DB support) are already implemented; the API/SDK/CLI are stable enough for pilot projects.
• **Risks** – License compliance, security hardening, and long‑term maintainer commitment still require a final review, but no major metadata or architectural red flags are present.

Overall, dense‑mem is a high‑readiness OSS component that can be introduced incrementally to give AI agents a robust, queryable memory layer and to standardize multi‑agent workflow orchestration.

Use in a workflow

markhuangai/dense-mem helps turn isolated prompts and tools into repeatable agent workflows.

• coordinate multi-agent workflows
• add tool-use pipelines
• standardize agent memory

Limitations and review notes

High for an OSS candidate: recent activity, adoption, and ecosystem signals are strong enough for a serious pilot.