Skip to content
cjjutbaPublic

About

A Mac app where Claude Code and Codex agents work as a team.

Topics

Resources

Contributing

Security policy

Stars

2 stars

Watchers

0 watching

Forks

Latest commit

 

History

1,096 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Kernel

Your AI coding agents, finally organized. Say what you need, and Kernel plans it, hands it out and brings it back for your approval. Kernel runs Claude Code and Codex agents on your Mac, each in its own git worktree. They ask before anything risky, open pull requests and wait for you to merge.

Download Kernel for Mac (Apple silicon) · Website

Projects: each project's agents and status, with its Lead chat and workspaces in the sidebar

Requirements

  • A Mac with Apple silicon, on a recent version of macOS
  • A Claude plan for Claude Code, a ChatGPT plan for Codex, or both. Either one is enough. Kernel brings its own copy of each.
  • git
  • The GitHub CLI (gh), signed in, for pull requests

On first launch Kernel checks Claude Code, Codex if a role runs on it, your sign-ins and the GitHub CLI, and tells you how to fix whatever is missing.

How it works

Projects and the team

Each repo you add is a project, and every agent in the repo's .claude/agents/ folder is on its team, along with your personal agents. Settings, Agents lists them and sets which ones take tasks. The sidebar lists each project's Lead chat and its workspaces, and the Inbox shows what needs you across all of them. Every status comes from real Claude Code and Codex events. Nothing changes on a timer, so an idle agent looks idle.

Settings, Agents: the roles and every agent on the project's team

Kernel plans and hands off

Kernel is the Lead. Open the project's Lead chat from the sidebar and brief Kernel in a sentence or two. The Lead runs on Claude Code or Codex, whichever you set in Settings, Models and effort. Kernel works out a plan in plan mode and brings it to you. You approve it or ask for changes. Once you approve, Kernel creates one workspace per task and hands each one to the right teammate. Each teammate runs on Claude Code or Codex, as set for its role.

Kernel's plan for four tasks, waiting for approval in the Lead chat

Workspaces and pull requests

A workspace is one task: a git worktree on its own branch, with its own port for the dev server. You can chat with the agent, read the diff file by file, run setup and run scripts, and open a pull request when it's ready. Checks show up in the workspace. After the merge, archive the workspace and it moves to History.

A workspace after its pull request merged

The Inbox

Anything that needs a decision lands in the Inbox: a command an agent wants to run, a plan to review, a question, a PR that's ready to merge. You can approve a command once or always allow it in that project. Nothing merges by itself.

The Inbox with a database command waiting for approval

Checkpoints and the big terminal

Kernel snapshots the worktree after every agent turn. If a turn goes wrong, use Revert files to here in the menu under an earlier reply and the files go back while the chat stays. When you want the plain CLI, open a terminal tab in the workspace. It runs claude or codex in the same worktree, and Kernel still follows the session.

The menu under a reply, with Revert files to here

A Claude Code terminal tab inside a workspace

Privacy

Kernel runs on your own Claude or ChatGPT plan, using the sign-in you already have. There is no Kernel account and no Kernel server. Your projects, chats, settings and history stay in a local database on your Mac.

Kernel itself makes these network calls and no others:

  • Claude Code and Codex, which talk to Anthropic and OpenAI as they always do
  • Anthropic or OpenAI, once, to check an API key when you add one in Settings, Accounts
  • gh, for pull requests and checks
  • git, to fetch and push branches on your repo's own remote, and your package manager when a new project installs dependencies
  • The update check, which reads this repo's GitHub Releases
  • An offline check, a DNS lookup of api.anthropic.com
  • Linear, only if you connect it in Settings
  • Usage data, only if you turn it on in Settings, Privacy. Kernel asks once. It sends anonymous counts of what goes wrong and the class of an error, never code, prompts, file paths or names, and Settings, Privacy lists every event it kept or sent

Build from source

You need Node 22 or later and the Xcode command line tools.

git clone https://github.com/cjjutba/kernel.git
cd kernel
pnpm install
pnpm dev

pnpm install also rebuilds the native modules for Electron. pnpm dist:mac builds an unsigned app into dist/. See CONTRIBUTING.md for tests and pull requests, and docs/RELEASING.md for signed releases.

How Kernel was built

Claude Code agents built Kernel one Linear issue at a time, working from a canvas of screen designs. docs/DECISIONS.md and docs/decisions/ record every choice someone might later want to undo, with the reason. They are kept as they were written.

License

Kernel's source is available under the Elastic License 2.0. In plain terms:

  • You can use Kernel for free, fork it, modify it and self-host it, for personal or commercial work.
  • You can't sell Kernel to others as a hosted or managed service.
  • You can't remove or work around license keys, if Kernel ever has them.

This summary isn't the license. LICENSE.md is.

Contributors sign the contributor agreement before their first pull request merges.

About

A Mac app where Claude Code and Codex agents work as a team.

Topics

Resources

Contributing

Security policy

Stars

2 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages