Why we built RepoOps
The launch essay. Why "your AI wrote the code, RepoOps remembers it" needed a product, not a blog post.
Every AI coding tool ships with a memory hole.
The IDE remembers the file. The agent remembers the message. Neither remembers why this codebase chose what it chose — the decision that landed two months ago, the anti-pattern that bit someone last quarter, the load-bearing one-line workaround whose original commit message just said wip.
RepoOps closes that hole.
The three pillars
RepoOps is built around three things every AI-built codebase needs and no AI tool gives you:
errors, sessions, files map. Read by Claude Code, Cursor, and Codex natively after npx repoops init.
per-developer rollups, prompt cadence. No upload, no API.
CLAUDE.md already remembers. The brain learns from outcomes.
- Memory. Flat-markdown brain files next to the code: decisions,
- Sight. Every Claude Code session captured locally — cost, tokens,
- Action. The brain proposes lessons; you accept; tomorrow your
Free for solo devs. Paid for teams.
The desktop app is free, OSS, and works locally without an account.
The hosted dashboard, team rollups, audit log, SAML/SCIM, and budget alerts are the paid surface. We charge for hosting + coordination, not for the moat.
Welcome.