Celis

Stop babysitting your coding agent

Celis keeps your rules and gives your agent the discipline to follow them, in every project and every agent.

https://github.com/user-attachments/assets/8cb59450-fb36-4fa6-82ae-5bcd59b1384b

Why

• Your agent follows through. It gets the rule that matters right before it acts, not buried in a manual it read an hour ago. You stop catching the same mistake twice.
• Correct it once. A correction in chat or a code review becomes a rule that Celis keeps, in every project. On a new job, your agent also starts from similar work that went well.
• Every agent keeps the same rules. What you tell Claude Code, Codex and Cursor follow too.
• You can check its work. Every step comes with proof, like a passing test or a command's output, on a trail you can watch.

How it works

A CLAUDE.md is a manual. Your agent reads it once at the start, and by the time it edits a file, the one rule that matters is lost among hundreds of others. Celis is a checklist. It hands your agent the few rules that apply right now: when a job starts, before it changes a file, and before it says it's done.

Those rules come from your corrections and code reviews. Your agent works one step at a time, tries the part it's least sure of first, proves each step works, and keeps what it learned as a rule. Each step lands on a trail you can watch, so you see what your agent did and why.

Job: seniors pay 20% less
├─ ✓ F