Navers lab
← Research

OpenChronicle

Local-first, open memory for tool-using AI agents

Open-source, local-first memory for any tool-capable LLM agent. Think OpenAI's Chronicle — but open, model-agnostic, inspectable, and hackable.

Navers lab · Einsia.AI · 2026

OpenChronicle

What it is

OpenChronicle runs on your machine, captures structured context from your real screen and app activity, and turns it into persistent, human-readable Markdown memory — files for projects, tools, topics, people, organizations, and daily events.

It is designed as a general memory layer rather than something tied to one model provider or app: run it fully locally, connect it to any model, and let any tool-calling agent read and build on it.

Highlights

Local-first

All memory stays on your machine — private by default.

Model-agnostic

Works with Ollama, OpenAI, Anthropic, or any LiteLLM provider.

AX Tree context

Uses the accessibility tree for efficient, low-cost context capture.

Inspectable

Human-readable Markdown memory, indexed with SQLite.

Tool-ready

Built for MCP clients and any tool-calling agent.

MIT licensed

Open-source and hackable, end to end.