Organised by what
you are doing
There is no separate manual. Each part of the site is the documentation for one job, and the deep reference lives with the code it describes.
Quickstart
Download, verify, start, load a dataset and run a query. Eight steps.
IntegrateBuild
The MCP block, the six agent tools, field types, and the HTTP API for everything else.
AskQuery syntax
Every operator with the field types it applies to, result shaping, sorting and what is refused.
Work with itClient CLI
The REPL, schema-aware completion, and ingest from files or URLs.
Run itOperate
Topologies, containers, security posture, keys and limits, and behaviour under load.
Get itDownloads
Binaries, deb and rpm, checksums, signatures and SBOMs.
The rest lives
with the code
Read the docs on GitHubVersioned with the release they describe
Configuration has several hundred settings and the architecture notes run to thousands of lines. Both change with the engine, so they are kept in the repository and versioned with it rather than copied here to go stale.
docs/CONFIGURATION.mdEvery setting: network, storage, search, cluster, security profiles, rate limits, audit, environment variables and multi-disk layout.docs/API_REFERENCE.mdRequest and response shapes for every route, with the capability each one needs.docs/ARCHITECTURE.mdThe actor hierarchy, the shard model and where blocking work is confined.docs/DEPLOYMENT.mdsystemd units, jemalloc tuning and multi-node topologies.crates/mcp/README.mdThe MCP server in full: tools, resources, transports and protocol revisions.