SwarnDB

SwarnDB is an AI developer tool and vector database that unifies vector similarity search with typed graph traversal in a single Rust-native engine, built for teams building GraphRAG, agent-memory, and recommendation systems. Its defining design is that a vector and a graph node are the same object — the vector ID equals the node ID — so there is no separate graph store to sync. The hybrid query engine runs vector search and directed graph traversal in one query plan, and edges carry confidence scores and provenance (manual or LLM-extracted, via a bring-your-own-key OpenAI-compatible model). Beyond retrieval, SwarnDB exposes 15+ server-side vector math operations including k-means, PCA, MMR, and analogy completion, so clustering and analytics run inside the engine instead of a separate pipeline. Performance comes from SIMD (AVX2/NEON), SQ8 quantization that cuts memory by roughly 75%, and HNSW indexing, with a published sub-7ms p95 latency at 99% recall on 1M vectors. It ships gRPC and REST APIs and deploys via Docker, a Python SDK, Kubernetes, and Helm. For anyone comparing vector databases for AI retrieval, SwarnDB targets the case where relationships between vectors matter as much as similarity.

Developers building GraphRAG, agent memory, and recommendation systems that need vector search and graph relationships in one engine.