Embedded retrieval database in Rust for mutable application data. Combines vector search, BM25 text search, metadata filtering, and hybrid ranking in a local store.
Work in progress. The Rust engine is implemented and under development. APIs and storage formats may change; this is not a production release. Mutable visibility, transaction rollback, recovery integrity and Python write acknowledgments are not yet qualified. Keep stored data rebuildable.
The omendb-vector-engine crate provides:
- Record insertion, replacement, deletion, and explicit transactions.
- Dense vector search, text search, and reciprocal-rank fusion for hybrid queries.
- Metadata filters and exact-search paths for checking retrieval results.
- WAL recovery, checkpoints, and persisted segments.
Canonical records own the stored data. Derived search indexes can be rebuilt from those records. See the architecture for the storage and retrieval contracts and current limitations. The evidence register contains external research inputs, not release qualification of the implemented engine.
Run from the repository root with Rust installed:
cargo build --workspace --all-targets
cargo test --workspace
cargo clippy --workspace --all-targets -- -D warnings
cargo fmt --all -- --checkThe Rust workspace currently contains the engine crate. Its public entry point
is Store, with APIs for records, transactions, search, and checkpointing.
Generate the API documentation with:
cargo doc --package omendb-vector-engine --no-deps --open