Cairn is an open-source distributed database for vector search, full-text search and filters, replicated with Raft and written in Rust.
It is built around one promise. Once a takedown is acknowledged, the document is gone from every replica, and a client holding the takedown's token will never read it again, through any node. That makes Cairn a fit for RAG over regulated or sensitive data, rights management, and anything where "we deleted it" has to be provable.
Where it stands: 0.3, developer preview.
- Scale: 50 million 128-dimension vectors on three 8-vCPU machines: filtered and unfiltered search at p99 of 31-36 ms, 435 queries per second, and a takedown visible on all three machines within 102 ms at p99.
- Wiring: your own ids, collections, tenants enforced by the API key, deletion by filter and by parent document, retention, partial updates, and signed proofs of deletion.
- Clients and integrations: TypeScript and Python clients, and vector stores for LangChain and LlamaIndex.
- Testing: thousands of simulated fault-injection runs with zero safety violations.
Every number comes with its report, including the misses.
Get involved. We are looking for:
- RAG and AI engineers who need retrieval they can trust;
- data-governance and compliance specialists who can say what proof of deletion should cover;
- systems engineers who like consensus, storage and search.
Start with the README, the
roadmap, and issues labelled
good first issue.
Contact: contact@cairn-db.com