Skip to content

Getting Started

Hippocampus is designed to stay local. You need a SQLite database and a local ONNX model path before the first write or search.

Install

bash
cargo install hcamp

Requirements

The model assets must be available locally.

  • model.onnx
  • tokenizer.json

Point --model-path at the .onnx file. The CLI accepts the flag before or after the subcommand.

If the configured model or tokenizer is missing, Hippocampus fails clearly instead of falling back to dummy embeddings.

Build

bash
cargo build --workspace

Test

bash
cargo test --workspace

CLI Usage

bash
hcamp write "Changed auth system" --event-type decision --tags auth,security --importance 0.9
hcamp search "auth"
hcamp recent 20
hcamp tags auth
hcamp reflect "why did we change auth system?"
hcamp read 7e866f1a-074c-4e1f-8c82-c38cc7ff04ac

Notes

  • The system is fully local and has no network or API layer.
  • Retrieval is deterministic.
  • Raw events are append-only and are never deleted by background processing.
  • The local embedder uses tract-onnx; incompatible exports are reported as structured embedding errors.

Example

bash
hcamp write "Switched to JWT for auth" --model-path ./model.onnx --db-path ./memory.db --event-type decision --tags auth,security --importance 0.9

The first run with --model-path stores the path in the database metadata so later runs can reuse it. :::