zinfradb
Database for AI apps
Start new thread
Zishan Neno

zinfradb - Database for AI apps

by•
zinfradb is a single-process database for AI apps: vector, graph, lexical, and hybrid search with durable jobs in one C++20 binary. Built for RAG, agent memory, and retrieval-shaped products. Fuses vector similarity, BM25 lexical, and graph evidence in one query with caller-tunable weights and per-result explain. Top of the ANN benchmark board at recall ≥ 0.95. One docker run, admin UI included. Source-available under BSL 1.1.

Add a comment

Replies

Best
Zishan Neno
Maker
📌
Building with AI, I kept hitting the same wall: models forget everything the moment a conversation ends. The usual fix — bolt on a vector database — papers over it, but a bag of embeddings isn't really memory. It finds text similar to a query, yet has no idea where a fact came from, what it contradicts, or what changed last week. I wanted memory that understands how knowledge connects, not just how it clusters. So vector, full-text, and graph live in one engine.