Einsia · 2026-04-25 · major
OpenChronicle — Open-Source Local-First Screen Memory for LLM Agents
Open-source MIT-licensed alternative to OpenAI's $100/mo Chronicle. Uses macOS accessibility tree events — not screenshots — to build Markdown + SQLite agent memory, then exposes an MCP endpoint so Claude Code, Codex, or any MCP-compatible agent can query persistent session context.
Free, local-first screen memory for AI agents — built in 48 hours as a direct response to OpenAI paywalling Chronicle at $100/month.
What is it?
OpenChronicle is a local-first agent memory layer for macOS. It watches your screen via macOS accessibility tree events (not raw screenshots), compresses activity into sessions, extracts durable facts, and stores everything as indexed Markdown files in a SQLite database. Any MCP-compatible agent — Claude Code, OpenAI Codex, Ollama-backed models — can query this memory through tool calls. MIT licensed, early alpha (v0.1.0), 1.1k GitHub stars within days of release.
How does it work?
Instead of capturing screenshots, OpenChronicle reads structured accessibility tree events from macOS, giving it app-level structured context without storing raw pixel data. Sessions get compressed and facts extracted by a local LLM (supports Ollama, OpenAI API, Anthropic API, or any LiteLLM provider). The built-in MCP server exposes memory as queryable tools, making persistent context available to any agent that speaks MCP.
Why does it matter?
OpenAI launched Chronicle gated behind a $100/month plan. Within 48 hours a team shipped a free, privacy-preserving, open-source equivalent. The accessibility-tree approach avoids storing screenshots — directly addressing the core privacy concern raised by Microsoft Recall and OpenAI's own Chronicle. Reaching 1.1k stars quickly signals real demand for an open alternative.
Who is it for?
Developers using Claude Code, Codex, or any MCP-compatible agent on macOS
Try it
github.com/Einsia/OpenChronicle