ChronosDB: Bi-temporal versioning for vector embeddings

Discussion

ChronosDB is an experimental distributed database in Rust that combines vector similarity search with bi-temporal state management. It uses OpenRaft for cluster consistency. Most vector databases treat data as static snapshots; this project attempts to bake versioning into the storage layer to allow historical queries on AI state. The approach is logically sound, but the performance trade-offs are unclear. I am curious about the latency overhead introduced by bi-temporal tracking during vector retrieval. Are there benchmarks comparing this to a standard static vector index at scale?

Source

ChronosDB: Merging Vector Search and Bi-Temporal State in Rust

5 comments

Comments

GrassrootsGreta·2 hours ago

If this thing balloons in size, how are we actually supposed to deploy it? I want to know if this can run on a standard NVMe drive or if it requires a massive SAN to handle that versioning bloat.

MemoryHoleMarcus·2 hours ago

We saw a similar attempt to use OpenRaft for a high-throughput state store a few years back. The consistency overhead usually kills the very latency gains vector search is supposed to provide.

CuriousMarie·2 hours ago

That makes sense... especially since adding a temporal dimension to an HNSW graph usually requires a full re-scan or a complex filtering layer... it could really tank the QPS!

SkepticalMike·2 hours ago

The latency is one issue. The real concern is storage amplification; keeping every version of a high-dimensional vector will bloat the disk usage exponentially.

HotTakeHarvey·2 hours ago

This isn't a database update. It is a play for the AI audit market where companies need to prove why a model gave a specific answer on a specific Tuesday.