v2.2.0 — works with 6 agents

Your AI coding configs
in one place.

Every AI coding tool has its own config format — your rules, skills, agents, and MCPs drift the moment you switch tools or add a developer. Codi manages all of them in one .codi/ folder and generates the right native config for every tool, automatically.

npm install -g codi-cli@latest

Your agents don't
share the same rules.

Every tool has its own config. Every new hire has to set up from scratch. Every teammate uses slightly different rules.

Without Codi

The problem with multiple AI tools

You add a security rule to Claude Code. Your teammate's Cursor doesn't know about it. Your CI runs Codex with whatever rules it had two weeks ago.

  • Every agent has a different config format
  • New teammate? Set up from scratch again
  • Rules drift the moment someone edits directly
  • No way to enforce team-wide standards
With Codi

Write once, works everywhere

Drop your rules in .codi/ and run codi generate. Every agent gets updated instantly, always in sync.

  • One folder drives all 6 agents
  • New teammate runs codi init and inherits everything
  • Git-tracked — every change is auditable
  • Lock rules your team can't override

All you need.
Nothing you don't.

Everything to keep your AI agents in sync — from first setup to team-wide enforcement.

One folder, 6 agents

Everything lives in .codi/. Codi reads it and writes the right config for each agent — Claude Code, Cursor, Codex, Windsurf, Cline, and GitHub Copilot.

Explore →

Don't start from scratch

There are 148+ templates already there for common languages, frameworks, security, and testing patterns. Pick what fits, skip what doesn't.

Explore →

Drift detection

If a generated file gets edited directly, Codi notices. You'll always know when something's out of sync.

Explore →

Share your setup

Pack your rules, skills, and agents into a preset and push it to GitHub. Everyone on your team installs it with one command — same config, same agent behavior, no setup time.

Explore →

Locked flags

Three layers: preset, repo, user. As a tech lead you can lock the rules your team can't change — things like security checks or test coverage requirements.

Explore →

14 hook types

Auto-detects Husky, pre-commit, and Lefthook. Secret scanning and type checking built in.

Explore →

Watch + revert

Live watch mode regenerates on every save. Roll back to any timestamped snapshot instantly.

Explore →

AI onboarding

codi onboard explores your codebase and proposes the ideal configuration automatically.

Explore →

Verification tokens

SHA-256 hash embedded in every generated file. codi verify confirms zero tampering.

Explore →

Works for solo devs
and teams alike

Same tool, different payoffs depending on whether you're working alone or with a team.

Solo Developer

Gets better the more you use it

Start from a preset. Each time an agent does something you don't like, write a rule. Over time your .codi/ becomes a pretty accurate picture of how you like to work — and any new AI tool you pick up just inherits it.

  • Running in 60 seconds with a built-in preset
  • Noticed a pattern the agent keeps missing? Write a rule.
  • codi onboard proposes the ideal config for your stack
# Sprint 1: start from a template
codi init
# Sprint 5: capture what you learned
codi add rule my-error-handling
✓ All agents updated with your conventions
# Sprint 20: agents that feel like you
Engineering Teams

Everyone works from the same rules

Right now every developer on your team probably has their own AI setup, and those configs don't talk to each other. Someone adds a rule, nobody else gets it. New hire joins — they start from zero. With Codi, the tech lead sets things up once and everyone installs the same preset. New hires are up and running in minutes.

  • Set it up once, share it with the whole team
  • Lock the rules that matter — security, testing, whatever your team needs
  • New hire runs codi init and gets everything
# Tech lead, once
codi contribute # publish team preset
# Every developer, including new hires
codi install github:acme/team-preset
✓ 12 rules, 8 skills — all agents synced
# One config. Zero drift. Always.

The more rules you write,
the better your agents get

You already know what good code looks like in your project. Your agents don't — not yet. Every rule you write is a bit of that knowledge going into Codi, and from that point on, every agent knows it too.

01

Code with your agents

Ship features. Notice where agents miss your conventions, reach for wrong patterns, or produce inconsistent output.

02

Write a rule

Takes a few minutes. Captures exactly what was off. From that point on, every agent follows it.

03

Regenerate and sync

Run codi generate. All 6 agents pick up the new rule right away.

04

Better output next sprint

The rule applies from the next prompt. That issue doesn't come back. Do it again — and your agents keep getting better.

Sprint 1 — Unconfigured agent
✗ Generic naming, no conventions
✗ No error handling style
✗ Ignores your test framework
✗ Missing security guardrails
✗ Inconsistent TypeScript patterns
Sprint 20 — Your tuned agent
✓ Follows your exact naming rules
✓ Applies your error handling style
✓ Writes Vitest tests your way
✓ OWASP checks on every suggestion
✓ Strict TypeScript, zero exceptions
148 Total artifacts
66 Skills
28 Rules
22 Agents
32 MCP servers
v2.2.0 MIT License

Up and running in 60 seconds.

Three commands. No manual config. Every agent in sync from the start.

1

Install

npm install -g codi-cli@latest
2

Initialize

codi init

Picks a preset, scaffolds .codi/, and generates all agent configs automatically.

3

Verify

codi status

After editing .codi/ files, run codi generate to resync.

bash
~ $ npm install -g codi-cli@latest
+ codi-cli@2.2.0 added 1 package
~ $ codi init
? Select preset balanced
.codi/ initialized
CLAUDE.md
.cursorrules
AGENTS.md
.windsurfrules
.clinerules
All agents synced. Zero drift.
Leandro A. Hidalgo

Leandro A. Hidalgo

The industry moves fast. I've gone through Claude Code, Codex, Windsurf, and Cursor — and my team kept adding new ones. Every agent had its own config, its own format, its own rules. And it wasn't just engineers — designers, product people, and business folks were using these tools too, each with completely different needs. Keeping everyone's knowledge in sync, across all these tools, was a real problem. That's why Codi exists — and why it keeps growing.

Questions? Say hello.

Have a question, feature request, or just want to talk AI agents?