RetrievalKit documentation
Start with the language you ship.
Each guide follows one path from package installation to a local query, explicit cleanup, and the platform boundary that applies today.
Language guides
Install, build, search, close.
The wrappers adapt types and concurrency; everything else stays in the Rust core.
Retrieval, graph, embedding, and pipeline products for Apple platforms.
macOS 14+ arm64 and iOS 15+ arm64PyPIPythonFlat retrieval, graph-scoped retrieval, and independent local embedding.
macOS arm64, CPython 3.10-3.14npm / Node.jsTypeScriptTyped asynchronous N-API wrappers for flat and graph-aware retrieval.
macOS arm64, Node.js 22.13+ or 24 LTSWorkers / WebAssemblyBrowserDedicated retrieval and embedding Workers with portable and SIMD128 tiers.
Published Worker packages; persistence deferredMaven CentralKotlinBlocking typed JNI for JVM, plus packaging-qualified Android artifacts.
JVM macOS arm64; Android API 24+ arm64-v8a previewChoose a query path
Choose how your app finds the right records.
Rank a corpus, return connected records, or let relationships narrow the corpus before ranking. Each path has its own API.
See how each path executes →RetrievalDatabase- Rank a flat corpusUse exact vector, BM25, or hybrid ranking when relationships do not define which records are eligible.
GraphDatabase- Follow relationshipsReturn records connected by application-supplied relationships, without vectors or an embedding model.
GraphRetrievalDatabase- Narrow, then rankLet relationships choose the eligible records, then rank only inside that scope with exact vector, BM25, or hybrid search.
Reference
Know what is local, measured, and supported.
Trace the execution model, inspect the evidence, confirm your target, or run the complete browser pipeline.