Einsia/OpenChronicle

OpenChronicle

Open-source, local-first memory for any tool-capable LLM agent.

Think OpenAI Chronicle - but open, model-agnostic, inspectable, and hackable.

Status: v0.1.0 · macOS only · early alpha

OpenChronicle gives AI agents a local, inspectable memory built from real screen and app context.

It runs on your Mac, captures structured context from what you're doing, and turns it into persistent Markdown memory: what you're working on, what you've decided, which tools you use, and which people or projects matter.

Any agent that can call tools can use it. MCP clients work especially well today, but OpenChronicle is meant to be a general memory layer for tool-using agents - not something tied to one protocol, one model provider, or one app.

Why OpenChronicle

OpenAI Chronicle points to an important future: agents that remember your real working context.

OpenChronicle is our open alternative:

• Local-first - memory stays on your machine
• Model-agnostic - use Ollama, LM Studio, OpenAI, Anthropic, or any LiteLLM-compatible provider
• Tool-friendly - usable by any tool-capable agent
• Inspectable - Markdown on disk, SQLite locally
• Open - MIT-licensed and built to be extended

Why AX-first

OpenChronicle currently prioritizes AX Tree / accessibility-tree context as its primary signal, with screenshots as a secondary signal over time.

We think this is the right tradeoff for an early memory system:

• Lower cost - structured text is far cheaper to process than screenshot-heavy OCR / vision pipelines
• Better intent capture - AX is often better for active app, focused element, edited text, URL, and interaction state
• Smaller, cleaner memory - easier to deduplicate, normalize, index, and retain long-term
• Better foundation - screenshots can later enrich visual context where AX falls short

AX-first for accurate, compact, low-cost memory; screenshot-assisted for richer multimodal context.

OpenChronicle vs OpenAI Chronicle

| | OpenAI Chronicle | OpenChronicle | | ------------------- | ------------------------------- | ---------------------------------------------- | | Source | Closed | MIT, open-source | | Model choice | OpenAI-centric | Your choice | | Who can use it | Product-specific workflow | Any tool-capable agent | | Primary capture | Screenshot / OCR-heavy | AX Tree first, screenshot-assisted | | Storage | Local generated memories | Markdown + SQLite on your machine | | Extensibility | Limited | Hackable parsers, memory logic, integrations |

How it works

flowchart LR
    W[mac-ax-watcherevents]
    S0["S0 dispatcherdedup · debouncemin-gap"]