About

The Mission

AI agents are only as good as their memory. Today that memory usually lives in a vector database with a fixed API: the agent can call semantic_search(query, k) and little else. Every custom retrieval strategy — recency weighting, importance filtering, cross-referencing — has to be reimplemented in application code, over and over.

We asked a simple question: what if the agent could write a program to query its own memory?

Liath is the answer. It is programmable memory for AI agents. The agent generates a sandboxed Lua program, and Liath runs it against a built-in key-value store, vector search, and embeddings. The LLM expresses the retrieval strategy directly — safely, with no file, network, or system access.

The SQLite for AI agents

We describe Liath as the SQLite for AI agents because it is embedded. There is no server to run, no cluster to operate, no vendor API to call. Liath ships as a single dependency and runs in your process, pointed at a data directory. That makes it a natural fit for local-first apps, edge deployments, and the many agents that are really just one process holding a working set of memories.

Built with Rust

Liath's core is written in Rust for memory safety, speed, and portability. Storage uses Fjall for the key-value layer, USearch for vector similarity, and FastEmbed for local text embeddings. On top of the Rust core, Liath ships as two editions of the same product: liath for Python (pip install liath) and liath-rs as a Rust crate.

The incredlabs data platform

Liath is built by incredlabs, founded by Dipankar Sarkar. It is part of the incredlabs data platform alongside ORMDB, a relational database for AI-era applications. Where ORMDB is the database for your application data, Liath is the memory for your agents.

Open Source

Liath is MIT licensed and developed in the open on GitHub. Current status: Alpha (v0.1.0). Contributions and feedback are welcome. Questions? Email contact@incredlabs.com.