OriginChain documentation.
OriginChain is a managed AI-native database. One endpoint, four query shapes, one substrate. SQL, vector, full-text, and graph all share the same hash-keyed k/v store, the same WAL, the same backups, the same single-tenant instance. Add an English /ask on top.
A row written once is visible to SQL, vector search, BM25 full-text, and graph traversal the moment the WAL fsyncs. No ETL, no second system, no eventual consistency between engines. Read the architecture overview, see the pricing tiers, or jump into the quickstart below.
Five minutes from signup to your first SQL, vector, full-text, and graph query. curl and Python side by side.
Pick a query shape.
All five share the same instance, the same bearer token, the same row store. A row written once is visible to SQL, vector, full-text, and graph the moment the WAL fsyncs.
SELECT · INSERT · DELETE · GROUP BY · HAVING · INNER · LEFT · RIGHT · FULL OUTER · 5-table JOIN POST /v1/tenants/:t/sql. Left-deep tree, projections, WHERE, LIMIT — full analytics surface, p99 < 8 ms on indexed reads.
HNSW · cosine · dot · L2 · filtered topk · SIMD f32 · tunable speed/recall HNSW with metadata filters and two modes. Default high_recall hits recall@10 = 0.96 at 100k with p99 109 ms; fast mode runs p99 37 ms at recall 0.69. f32 SIMD distance kernels.
boolean AND · BM25 · phrase · 18-language stemming · UAX #29 Per-field, per-tenant inverted indexes. Lucene defaults (k1=1.2, b=0.75). Diacritic folding and stop-words optional.
neighbors · reverse · bfs · path · dijkstra Direction-tagged relation keys. BFS up to max_depth and weighted shortest path with caller-supplied weights.
POST /v1/ask · plan-cached · Bedrock · Anthropic · OpenAI An English sentence becomes a Plan tree. First call compiles, every repeat is a cache hit at p50 < 2 ms.
Or by topic.
The substrate, the SDK, the dashboard walkthroughs, the deployment story, and the operational surface — five doors into the same engine.
The substrate, key shapes, schemas, the Plan tree, replication model, versioning, and PITR.
Typed Python client. Sync and async. Pandas-friendly row coercion. Generated from your schema catalog.
Step-by-step guides for inserts, schemas, queries, backups, and restoration in the managed console.
Console provisioning, region pinning, single-tenant EC2, sync follower, ACM TLS, Route 53.
Backups, sealed-segment shipping, point-in-time recovery, Prometheus metrics, audit log, status page.
Every HTTP endpoint — schemas, rows, sql, vector, fts, graph, ask, watch, ops. Bearer auth, idempotency keys, rate-limit headers.
Machine-readable spec at /openapi.json. Feed to any SDK generator or AI coding agent — they'll regenerate clients in any language.
@originchain/mcp-server — drop into claude_desktop_config.json and your AI agent can ask, SQL, vector-search, FTS, and list schemas on your tenant.