Keel#
One config, every AI agent, governed from day one.#
Your AI agent config is scattered across projects, duplicated between teams, and impossible to enforce. Keel compiles layered configuration into any agent’s native format — Claude Code, Cursor, and beyond.
One Command to a Configured Agent#
Pull a community config and compile — Keel detects your agent automatically:
keel add 'https://github.com/affaan-m/everything-claude-code.git#main'Stack your own config on top:
keel add ./my-toolsEvery keel add updates .keel/config.yml and recompiles.
The result is a managed config that your whole team can use:
# .keel/config.yml
layers:
- 'https://github.com/affaan-m/everything-claude-code.git#main'
- ./my-tools
output:
claude: .keel compile # repeatable builds from configSame layers, different agents — add outputs:
keel compile --output claude=. --output cursor=.What Keel Does#
Community#
Pull skills, commands, and configs from any Git repo or HTTP archive.
Existing repos with .claude/ or .cursor/ work as layers with no changes by the author.
Cross-Agent#
Compile the same config to Claude Code, Cursor, and more. Switch agents without rewriting your setup. Native config works as a layer — no migration required.
Layers#
Stack configuration from community → company → team → project → personal. Each layer adds or overrides the layers before it. Ship scripts and binary dependencies alongside your config.
Import#
Already have Claude or Cursor config? Import it as a Keel layer in one command.
Keel reads your native config and produces a keel.yml with companion files.
Templating#
Variables, conditionals, and reusable snippets built into the compiler. A/B test prompts, adapt instructions per-agent, create parameterized skill templates — all at compile time.
Governance#
Higher-trust layers lock values that downstream layers can’t override. Force-lock security hooks, mandate MCP servers, enforce permissions. Lockfile pins layer hashes for reproducible builds.
Use Cases#
Personal Toolkit#
Assemble your global agent config from community repos and personal tools in one command.
Multi-Agent Portability#
Write config once, compile to Claude Code and Cursor from the same source.
Enterprise Policy#
Enforce security hooks, mandate MCP servers, and lock permissions across every team.
Developer Onboarding#
New developer joins, runs keel add, gets the full agent environment — instructions, MCP servers, permissions, skills, tools.
From Personal Toolkit to Enterprise Policy#
Keel scales with you. Start by assembling your personal config from community sources. When your team needs consistency, add a shared layer. When your org needs governance, add force-locks and policies.
Individual Developer#
Assemble your agent config from community repos and personal tools.
keel add installs each layer and compiles in one step.
~/.claude/
├── CLAUDE.md ← community + personal instructions
├── settings.json ← your model preferences
├── commands/ ← community + your commands
├── skills/ ← community skills
└── agents/ ← community agentsTeam Lead#
Define team conventions once, compile into every project.
New developers run keel compile and get the full setup.
# .keel/config.yml
layers:
- ./team-config
- ./project
output:
claude: .Platform Engineer#
Enforce company-wide security policies across all teams. Force-lock hooks, mandate scanners, lock down permissions.
# .keel/config.yml
layers:
- ./company
- ./team
- ./project
output:
claude: .
cursor: .Zero Risk to Try#
Keel compiles into your existing project — it doesn’t replace your config, it adds to it.
Your manually created files are never touched.
keel clean removes only what Keel created, leaving everything else intact.
Already have a .claude/ or .cursor/ directory?
It works as a Keel layer as-is — no migration, no conversion, no lock-in.
Open Source, Self-Hosted, No Vendor Lock-In#
Keel is a CLI compiler — not a platform, not a SaaS, not an agent. It works with any AI coding agent that reads config from the filesystem. Your config stays in Git, follows your PR process, and runs wherever you do.
The keel is the structural backbone of a ship — everything else is built on top of it, it keeps the vessel stable, and it’s invisible from above. That’s this project: the infrastructure underneath the AI agent that provides structure, governance, and stability, but the agent never knows it’s there. It just sees the same native config files it always has.