Skip to content
Geordi Features Install GitHub Download

A calmer way to run coding agents.

Like other agent orchestrators, Geordi gives every task its own agent, Git worktree and branch. The difference is in what happens around them: your agents keep working after you quit, they can bring in other agents to help, and every workspace shows you the next step towards a merged pull request.

Download Geordi

Version 0.7.0 for macOS (Apple silicon), Windows and Linux. It's still early days, so expect things to change.

Return people after sign-in

After signing in, people land on the home page instead of the page they came for. Send them back there.

Claude Code

People now return to the page they asked for. The sign-in form carries it as returnTo, and only paths on this site are accepted. Tests pass.

Claude CodeClaude

Follow a fix from the first message to a merged pull request. Replies stay in this demo.

What Geordi does differently

Quitting doesn't stop your agents

Your agents run in a background process of their own, not inside the window. Quit Geordi or install an update in the middle of a task and they carry on. When you open it again, everything they did while you were away is there.

You quit Geordi18:02

While Geordi was closed

  • Claude finished, tests passreturn-after-sign-in18:09
  • Codex opened a pull requestaudit-log-pages18:17
  • Claude has a question for youvite-818:31

You open it again, and it's all there18:40

Every workspace knows its next step

Geordi follows the work all the way to a merged pull request. The bar at the top of a workspace says where it stands and offers the next step: create the PR, fix failing checks, resolve conflicts, merge. The sidebar shows the same for every workspace, so you see at a glance which one needs you.

  1. 3 files changedCreate PR
  2. Checks failingFix errors
  3. ConflictsResolve conflicts
  4. Ready to mergeMerge PR
  5. MergedArchive

Agents can run agents

Ask Claude to have Codex review its work, or to split a job between a few helpers. Geordi gives every agent tools to start, message and stop other agents. The helpers aren't hidden away: each is a conversation of its own that you can read and step into.

Claude

Review the redirect

Find other redirects

I've asked Codex to review the change, and another agent to look for other places that send people to the home page. I'll let you know what they find.

Five dev servers, no port clashes

Tell Geordi in a .geordi.yml what your project needs to run. Every workspace then gets a Run button, terminals, and ports of its own that stay the same for as long as it lives. The same dev server can run in five workspaces at once, and the agents know their ports too, so they can start and check things themselves.

Read the configuration docs

.geordi.yml

setup: pnpm install

jobs:
  db:
    command: docker compose up db
  web:
    command: pnpm dev --port $PORT
    needs: [db]

urls:
  app:
    url: http://localhost:{web}/

A few things we decided on purpose

A quiet screen
Most of the window is the conversation and the diff. There is very little colour, and the only things that move are agents at work and checks that are running.
No plan mode
If you'd like a plan first, ask for one and tell the agent to wait until you've read it. While it works, Geordi shows you its to-do list.
Two permission modes
Full Access or Auto Mode, chosen per workspace. There's no mode where you have to approve every single tool call.
No built-in browser
Previews of your project, and any links, open in the browser you already use.
Claude and Codex, side by side
Pick the agent, model and effort for each workspace, and mix them in one project. Geordi runs the agents you already have.
It runs on your machine
There's no account and no cloud service. Your history is kept in a SQLite file in your home directory.

Try it

The desktop app comes with everything it needs and keeps its data in ~/.geordi. Geordi is young and changing quickly, so you'll probably run into a rough edge or two.

One thing to know: Geordi doesn't have logins yet, so it only accepts connections from your own computer. Agents run with Full Access by default, which means they can do anything your user account can. Choose Auto Mode if you'd rather keep them on a shorter leash.

  1. 1

    Sign in to Claude Code or Codex

    Geordi doesn't come with agents of its own; it uses the ones you already have. For GitHub, it uses your gh CLI login.

  2. 2

    Install the app

    On a Mac, open the disk image and drag Geordi into your Applications folder. On Windows, unzip it and run Geordi. On Linux, install the .deb package, make the AppImage executable and run it, or unpack the tarball.

  3. 3

    Start your first workspace

    Add a repository, describe what you want done, pick an agent and press send.