Free & open source, local app
In the spirit of herdr, Ogden Agents keeps many projects and many chats in one place and shows which are working, waiting on you, or idle. It wraps Claude Code, Antigravity, Codex and Grok through the Agent Client Protocol, plus local models, so you stop juggling a terminal tab per agent. Each one asks before it runs a command or edits a file, and you answer with a card.
What's inside
All of it is in one codebase you can read, change, and run on your own computer.
Use Claude Code, Antigravity, Codex, Grok, or a local model, each with its own persistent chats and model choice. The sidebar shows at a glance which are working, waiting on you, or idle, and you get a preview before handing a conversation to another agent.
Each agent asks before it runs a command or edits a file, and you answer with a card. Nothing happens unattended unless you choose Auto for an agent that supports it.
Turn on BMad Planning, a live ticket Board, and Retrospectives separately for each project, so a project only gets as much process as it needs.
Choose the Ogden, Forest, or Ember appearance preset, or build a fully custom theme with paired light/dark colors, contrast validation, and JSON import/export to share it.
Tickets build in their own worktree with the agent you pick: Claude Code, Codex, Grok or Antigravity. Each one says up front whether it builds on its own or with you watching. Inspect the verification results and approve the merge yourself. A blocked build says why, with Retry or Build again with another agent.
Coordinate a manager and worker agents from Orchestrate, with approval per instruction by default, recorded activity, and a Stop you can hit any time.
Turn on Developer mode for a Terminals tab per project: tabs and splits running a plain shell or the agents' own command line programs. Each terminal shows a best guess of what it's doing and can notify you when it may need you.
Connect MCP servers once in Settings and every coding agent gets them in every project. Keep reusable skills, named instruction snippets, the same way. When one agent runs out of usage, continue the same chat with another from an editable summary.
Run it
Ogden Agents is for one user on one computer: the server only listens on 127.0.0.1,
there's no telemetry, and nothing leaves your machine except what you explicitly send to an agent
you've signed in to.
No Node and no terminal. Download the file for your computer from the
latest release:
a universal .dmg for Mac, an x64 or ARM setup.exe for Windows, and an
.AppImage or .deb for Linux. The apps aren't signed yet, so your computer
warns the first time; the setup guide shows how to open them.
Clone the repo, install, and start it. The app opens in your browser at a one-time launch link.
$ git clone https://github.com/hsmith-dev/ogden-agents.git $ cd ogden-agents && pnpm install $ pnpm start
On the way
Being built in the open on GitHub. None of this is in a downloadable release yet.
Run an agent on another computer you own over SSH. Ogden makes a key for each machine and asks you to confirm its host key once. The app itself still only listens on your own computer.
Add any agent that speaks the Agent Client Protocol by its command, including one you wrote yourself, and chat with it like the built-in ones. Every tool call asks you first. The Local model can build tickets with you watching.
License
No tiers, no checkout, no seat count, no license key. Ogden Agents is MIT licensed — read it, change it, and run it wherever you like.
Clone the repository and read the source, change it, or ship your own fork — the MIT license covers personal use, work use, and everything in between on the same terms. Provider subscriptions and API usage for the agents you connect are your own cost; the app itself is free. It hasn't had an outside security audit, and it's built for one user on one computer.
It needs Node.js and an agent you can sign in to. Clone and run it in a few minutes.