MinnsDB
A single-binary Rust database purpose-built for AI agent workloads, combining a temporal knowledge graph, a relational table engine, and sandboxed WASM agents behind one query language.
MinnsDB combines a temporal knowledge graph, a bi-temporal relational table engine, and a sandboxed WASM agent runtime in a single Rust binary. Sending it raw chat transcripts triggers a multi-call LLM cascade that extracts entities and relationships and writes them to the graph as temporal edges with automatic supersession and cascade invalidation, without the developer writing extraction code.
Every graph edge carries valid_from and valid_until timestamps, so nothing is deleted, only superseded, and a query can ask what is true now, what was true at a past point in time, or what the database believed at a past moment. Tables use page-based storage with blake3 checksums and support SQL-style CRUD, joins, and graph-to-table joins.
Its query language, MinnsQL, covers graph pattern matching, table queries, DDL, DML, temporal clauses, and reactive subscriptions in one syntax. Sandboxed WASM agents can read and write tables, query the graph, call external APIs, and trigger on events, with instruction metering and memory and permission limits. It can run from source with Cargo or as a Docker image.