Show HN: AgentKV – SQLite for AI agent memory (MMAP vector+graph DB)

**Show HN: AgentKV – SQLite for AI agent memory (MMAP vector+graph DB)**

I built AgentKV — an embeddable vector + graph database for AI agents.The problem: I was building a local LLM assistant and needed to store conversation history + RAG documents. ChromaDB/Pinecone require running servers. FAISS has no persistence. I wanted something that just works like SQLite.AgentKV is a single-file mmap database with: • Vector search (HNSW index) + graph relations • Crash recovery with CRC-32 checksums • Zero servers, zero config files • Thread-safe concurrent reads • pip install agentkv (Python 3.9+) Real examples in the repo: • Local RAG with Ollama (10 lines) • Multi-turn chatbot with memory • Agent coordination with context graphsBuilt in C++20 with mmap + nanobind Python bindings. Benchmarked against FAISS — competitive throughput with built-in persistence.Open to feedback on the API design and any use cases I'm missing!PyPI: https://pypi.org/project/agentkv/ Comments URL: https://news.ycombinator.com/item?id=47027887 Points: 1 # Comments: 0

---

**[devsupporter 해설]**

이 기사는 Show HN에서 제공하는 최신 개발 동향입니다. 관련 도구나 기술에 대해 더 알아보시려면 원본 링크를 참고하세요.