Hello, CommitMind
title: "Hello, CommitMind" description: "A note on why we're building project memory for AI coding agents — and what ships first." date: "2026-04-24" author: "The CommitMind team"
Welcome. This is the blog's first post — the seed we keep around so the dynamic MDX route always has something to resolve. Real posts will replace it soon.
A short pitch while you're here.
Why project memory matters
Every AI coding session starts cold. Your agent doesn't know why you picked OAuth 2.1 over the legacy flow, which module became the boundary you promised not to cross, or that the "clever" ledger writer is actually the path of record. So it makes the wrong choice. You paste the README. You try again. You spend tokens teaching it things your team already decided.
CommitMind fixes this by capturing decisions where they happen — in your git history — and making them instantly available to any MCP-compatible client.
What's shipping now
- Commits, observations, and decisions, anchored to SHAs
- Hybrid search (FTS + vector) across your project's memory
- An MCP server that Claude Code picks up automatically after
commitmind init - Team features: invites, roles, a shared project graph
Check the changelog for the full list of what's shipped, and the quickstart to get going in under 3 minutes.