Feature details | Alopex DB

The architecture is on the feature overview.

Storage

All data goes into the same underlying storage. By default, writes are synced to disk in batches, so large volumes of writes finish quickly. The database recovers after a crash. For columnar tables, you choose per table between Zstd compression and no compression.

A database is a single .alopex file. Copy it to back it up or hand it over, and it reads as is at the destination.

Transactions

ACID transactions, with reads that never block writes.

In SQL, use BEGIN, COMMIT, ROLLBACK, and SAVEPOINT. The isolation level is REPEATABLE READ. READ ONLY transactions are available too.

SQL

JOIN, GROUP BY, recursive CTEs, and window functions
Constraints such as primary keys, foreign keys, and NOT NULL
Views
Full-text search
Execution plans with EXPLAIN
Export to CSV

For SQL examples, see the quickstart.

Vector search

Store embedding vectors in vector columns. Choose the distance from cosine, L2, and inner product.

Without an index, exact search scans every row and returns precise results. Build an HNSW index and search runs fast and approximate. Either way, combine it with SQL conditions to find the nearest rows among those that match.

Embedded, CLI, and server

Embedded: use it as a library from Rust and Python, in memory or on disk.
CLI: work with SQL, KV, and vectors from the command line.
Server: connect over HTTP and gRPC, with TLS, audit logs, and metrics.

Cluster foundation

Cluster state management and routing diagnostics. It behaves like a single node and does not affect ordinary SQL execution. Node-to-node communication uses the related product Chirps.