01
Quick start
Go from startup to write, search, filter, delete, and restart recovery in 30 minutes; covers Docker, the macOS prebuilt package, and source builds.
Documentation / Index
Docs that start with a path that runs.
Run it first, connect an SDK next, then make it durable with the deployment guide. Every entry points at the main branch of the GitHub repository; docs and code evolve together.
02
Deployment
Docker Compose, systemd, and Helm; ports, WAL directory, backups, and the boundaries of manual HA. Production must set --wal-path explicitly; the temporary default cannot carry production data.
03
Server manual
Startup flags, pgwire and HTTP APIs, the OpenAI-compatible gateway, metrics, and observability; includes HNSW tuning knobs and fallback switches. The runtime OpenAPI document is at /v1/openapi.json.
04
Embedding
The kernel does not run embedding inference; an OpenAI-compatible /v1/embeddings gateway forwards to OpenAI, Ollama, TEI, vLLM, or another upstream, and can auto-upsert with the collection field.
05
SDKs
Python, TypeScript, Go, Rust, .NET, and Java clients; Python also ships LangChain and LlamaIndex adapters. All six languages share the same stable_id rules.
06
MCP server
Source is ready: a stdio MCP server with collection listing, vector search, hybrid search, and opt-in writes. Run it from source until the npm package is published; the npx command will follow.
mcp/houyidb-mcp ↗Source available
07
Measurements and protocol
October 2026 single-machine runs: 1×/2×, contiguous-vector filter cache, and streaming plan/apply; includes scope, reproduction commands, and the measurement protocol. HouyiDB data only.