Ketebe documentation

Build a retrieval layer that owns ingestion, hybrid search, embedding lifecycle, governance, and agent access without stitching together a stack of sidecars.

v0.9 release preparation: the core product surface is implemented and tested. Packaged container and Docker Compose artifacts are being finalized, so the docs distinguish current source-build capabilities from the target packaged quickstart.

Start here

10 minute pathQuickstart

Create a collection, ingest a document, and run a hybrid query using the real Ketebe SDK surface.

Choose your pathGet started

Local evaluation today, packaged v0.9 onboarding next, and the paths toward production.

Release readinessRoadmap

See what v0.9 is finishing and what evidence Ketebe requires before calling v1.0 production ready.

Evidence over claimsBenchmarks

Understand the methodology Ketebe uses for repeatable retrieval, ingestion, recovery, and scale measurements.

Retrieval

Ketebe exposes dense vector, sparse and lexical retrieval behind one query surface. Query profiles can combine metadata filters, fusion, reranking, and explainability instead of forcing applications to orchestrate independent search systems.

result = client.query(
    "product-docs",
    QueryRequest(
        text="How does recovery work?",
        top_k=5,
        search_profile="balanced",
        explain=True,
    ),
)

Ingestion

Write records directly, ingest documents for server-side embedding, or feed continuously changing data through Kafka-native ingestion. Async jobs provide a product-level lifecycle for work that should not be hidden in application workers.

Embeddings & model lifecycle

Embedding is treated as managed lifecycle state. Ketebe owns server-side embedding and re-embedding workflows so model changes do not require every application to invent its own migration pipeline.

Agents & MCP

The first-party MCP adapter sits on the stable product API. It supports discovery, retrieval, context assembly, provenance and controlled ingestion without bypassing authentication, authorization, storage, or correctness boundaries.

Architecture & operations

Ketebe separates durable source-of-truth state from derived indexes. WAL-backed storage, recoverable segments, backup and restore, TLS/mTLS, data-plane isolation and operational contracts are part of the architecture rather than deployment folklore.

Security & governance

Organizations, projects, API keys, RBAC, quotas, audit events and isolation boundaries are built into the product surface. Authentication and authorization are modeled separately so security policy remains explicit.