The Architecture of Hippocampus

Hippocampus was built to solve a simple problem: agents forget things. We needed a local-first memory kernel that wouldn’t rely on expensive, latency-heavy cloud setups.

Why SQLite?

It’s universally available, incredibly robust, and with FTS5, it offers extremely capable keyword search. When paired with local ONNX embeddings, you get a powerful hybrid search setup that is entirely deterministic.

SDK First

We built this as a Rust SDK from day one to ensure maximum portability and performance.

The Future

We’re excited to see what developers build on top of our primitives. The tools developers need most, and agents want to use most.