Skip to main content
ConareDB standalone is in controlled beta. The engine backs Conare AI Memory in production; direct database access is being opened to beta customers. Live status: conare.betteruptime.com.
ConareDB is an object-storage-native search engine: durable local WAL plus an object-store mirror, hybrid retrieval (vector ANN + BM25 keyword + attribute ranking), typed filters, and binary bulk ingest — organized into fully isolated namespaces.

ConareDB Console

Sign in with your cdb_ key to browse namespaces, usage, and metrics.

​
Two ways to buy the same engine

Standalone ConareDB never turns text into embeddings — pair it with the Embeddings API if you want vectors from the same model family that powers Conare retrieval, or bring any model you like.

​
What you get

  • Isolated namespaces — each a complete store with its own WAL, persister, and object-store mirror. Created on first write; permanently deleted with a durable anti-restore tombstone.
  • Hybrid search — vector, BM25 keyword, and chronological branches in one request, with a typed filter DSL. Vector scores are always exact: the ANN path reranks candidates against full-precision vectors before returning.
  • Scoped keys — cdb_... bearers scoped to exact or prefix namespace patterns, with independent read / write / delete permissions, optional per-key rate limits and expiry. Secrets are shown once and stored only as SHA-256 digests.
  • Binary bulk ingest — one binary fp16 frame per request with restart-safe conditional receipts, built for seeding tens of millions of rows.
  • Deterministic results — for a fixed index state, identical queries return byte-identical hit lists at any concurrency. Continuous recall sampling checks ANN answers against brute-force ground truth in the background.

​
Endpoints at a glance

Start with the quickstart, or jump to search, writes & ingest, namespaces & keys, or migrating from Turbopuffer.