Introducing incredlabs: the data layer for AI-native software

Most of today’s data infrastructure was designed for a pre-AI world. Relational databases assume an application server translating objects into SQL. Vector databases assume a fixed similarity-search API. Neither was built for software where models, embeddings, and autonomous agents are first-class citizens.

incredlabs builds the data layer for AI-native software — open source, written in Rust, and designed around the primitives modern applications actually use.

Two products, one philosophy

ORMDB — the database that speaks ORM natively

Object-relational mappers spend enormous effort translating object graphs into SQL, which is where the classic N+1 query problem comes from. ORMDB flips this around: the database understands your entities, relations, and constraints natively, so you fetch an entire object graph in a single round-trip. It ships with native relations, ACID transactions, safe online migrations, change streams, and built-in vector, geo, and full-text search.

Liath — the SQLite for AI agents

Agents don’t want a fixed search(query, k) API; they want to write their own retrieval logic. Liath is programmable memory: agents write sandboxed Lua to filter, rank, and cross-reference their own memory, safely — with no file, network, or system access. It’s embedded and zero-infrastructure, with a KV store, vector search, and embeddings built in. Available for Python (pip install liath) and Rust (liath-rs).

Principles

• Rust all the way down for predictable performance and portability.
• Open source under the MIT License, developed in the open at github.com/incredlabs.
• Built for AI, with vector search, embeddings, and agent memory as first-class concerns.
• No lock-in. Run it locally, embed it, or deploy it anywhere.

This is just the beginning of the platform. If you’re building AI-native software and the data layer is fighting you, we’d love to hear from you: contact@incredlabs.com.