personal-model

Why we included this project

Coding agents tend to forget what they learned about you between sessions, and personal-model is built around that exact problem: it gives the agent a record it can keep working from. It runs locally on your Mac and, once you grant macOS permission, builds a persistent model of you from the focused activity it observes. That model is exposed through a single MCP server, so Claude Code, Codex, Cursor Agent, or any other trusted MCP client sees the same picture of your priorities and working patterns. The part that separates it from a plain chat-history dump is the evidence: every claim links back to the activity that produced it, so an inference can be corrected or overturned as new information arrives instead of lingering unnoticed. Developers who want agents to resume unfinished work and rank next steps against real priorities will appreciate that the memory stays on your machine, where you can inspect, correct, export, or delete it.