A lightweight SQLite planning and memory backend for local LLM agents.
Built for OpenClaw agents and designed to improve reliability when running smaller local models over long tasks.
OpenClaw Store gives an agent persistent state for:
- goals
- tasks
- task dependencies
- progress tracking
- notes
- long-term memory
- important facts and decisions
Instead of relying only on the model's context window, the agent can store and retrieve structured information across sessions.
Smaller local models are becoming increasingly capable, but long-running agent workflows often fail because the model loses track of:
- what it was trying to accomplish
- what has already been completed
- important decisions made earlier
- previous errors and lessons learned
- what should happen next
OpenClaw Store acts as an external memory and planning layer.
The model does the reasoning.
The store keeps the state.
This project is currently used with:
OpenClaw
and tested using:
Ornith-1.0-9B-Uncensored-Abliterated-GGUF
Model: https://huggingface.co/PeppX/Ornith-1.0-9B-Uncensored-Abliterated-GGUF
The goal of this project is to make smaller local models more effective by giving them a reliable external planning and memory system.