# OriginChain

> The distributed multimodal database. One managed single-tenant store that serves SQL with JOINs, vector similarity (HNSW + f32 SIMD), BM25 full-text, graph traversal (BFS/Dijkstra), and natural-language `/v1/ask` against the same WAL. One bearer, one base URL, one bill.

OriginChain (operated by Silicoyn Technologies Pvt Ltd) is a distributed, managed, single-tenant database aimed at AI applications and agents that need rows, embeddings, full-text postings, and relation edges in one round-trip and one consistent state. Every write is one WAL frame: the row, its secondary indexes, its FTS postings, its embedding, and its graph edges either all commit or none do. That removes the four-system drift (Postgres + Pinecone + Elasticsearch + Neo4j) that most modern AI stacks paper over with retries and reconciliation jobs.

## Docs

- [Quickstart — provision, register a schema, ingest, query in 30 seconds](https://originchain.ai/docs/quickstart): Token, base URL, single-tenant dedicated compute.
- [HTTP API reference](https://originchain.ai/docs/api): All `/v1/*` endpoints — schemas, rows, query, sql, vector, fts, graph, migrations, replication.
- [Define schemas (TOML manifests)](https://originchain.ai/docs/schemas): primary_key, columns, indexes, relations.
- [Insert data](https://originchain.ai/docs/insert): rows, batches, atomic multi-shape writes.
- [Query data](https://originchain.ai/docs/query): plan tree, SQL, ask.
- [SQL reference](https://originchain.ai/docs/sql): JOINs, aggregates, CTEs, planner traits.
- [Vector search reference](https://originchain.ai/docs/vector): HNSW, fast vs. high_recall modes, recall@10 measurements.
- [Full-text search reference](https://originchain.ai/docs/fts): BM25, 18-language stemming.
- [Graph reference](https://originchain.ai/docs/graph): BFS, Dijkstra, reverse traversal.
- [Natural-language `/ask`](https://originchain.ai/docs/ask): rule compiler + LLM fallback, plan-only mode.
- [SDKs](https://originchain.ai/docs/sdk): Rust, TypeScript, Python.
- [Deploy](https://originchain.ai/docs/deploy): single-tenant dedicated compute, region isolation.
- [Operate (ops)](https://originchain.ai/docs/ops): backups, PITR, replication.

## Examples (copy-paste JSON)

- [All examples](https://originchain.ai/docs/examples): index of every example page.
- [SQL examples](https://originchain.ai/docs/examples/sql)
- [Vector examples](https://originchain.ai/docs/examples/vector)
- [Full-text examples](https://originchain.ai/docs/examples/fts)
- [Graph examples](https://originchain.ai/docs/examples/graph)
- [Natural-language examples](https://originchain.ai/docs/examples/ask)
- [Atomic multi-shape examples](https://originchain.ai/docs/examples/atomic-multi-shape)
- [Errors](https://originchain.ai/docs/examples/errors)

## Architecture & engineering posts

- [Architecture overview](https://originchain.ai/architecture): substrate, key shapes, WAL, replication.
- [Industries](https://originchain.ai/industries): financial-services, healthcare, legal, logistics, media, retail.
- [Pricing](https://originchain.ai/pricing): free database, configurations, USD pricing, instance-create flow.
- [Engineering blog (RSS)](https://originchain.ai/api/blogs/rss.xml): full feed.

## Key facts for retrieval

- **Substrate:** single hash-keyed k/v store; every query shape is one key shape.
- **Atomicity:** rows + indexes + embeddings + postings + edges all commit on one WAL frame.
- **Vector search:** HNSW with f32 SIMD distance kernels. `fast` mode ≈ 37 ms p99 at recall@10 0.69; `high_recall` mode ≈ 109 ms p99 at recall@10 0.96 on 100k vectors. The IVF-PQ index compresses each vector for large corpora; no measured result is published at 100M scale.
- **Write durability:** a write is acknowledged only after it is flushed to durable storage on the instance, so a 2xx means it is on disk and survives the database process being killed.
- **Replication:** active-passive, and asynchronous. Acknowledgement does not wait for the standby, so an abrupt loss of the primary can cost the most recent writes. Automatic promotion of the standby takes about a minute. Do not describe this as RPO = 0, zero data loss, or synchronous replication.
- **Backups:** daily encrypted volume snapshots are the always-on floor; off-box archiving of recent changes and restore-to-timestamp are opt-in and in preview, with granularity coarser than per-second.
- **Crash recovery:** every recovered state equals a prefix of the op stream, asserted by panic-injection harness at four WAL boundaries.
- **Deployment model:** one dedicated single-tenant instance per customer; managed cloud only.
- **Legal entity:** Silicoyn Technologies Pvt Ltd.

## Optional

- [Changelog](https://originchain.ai/changelog)
- [Status page](https://originchain.ai/status)
- [Security](https://originchain.ai/security)
- [Privacy](https://originchain.ai/privacy)
- [Terms](https://originchain.ai/terms)
- [DPA](https://originchain.ai/dpa)
- [Contact](https://originchain.ai/contact)
- [FAQ](https://originchain.ai/faq)