huddle

Your whole team’s AI coding agents, working on one codebase at the same time.

You use Claude Code, a teammate uses Codex, someone else is on Cursor. Normally those three agents edit the same files on three laptops and you find out at merge time. huddle connects them. A file one agent is editing is locked for the others, edits show up on everyone’s machine as they happen, and the agents talk to each other to split the work.

It’s made for small teams where everyone brings their own subscription: hackathon teams, class projects, a few friends shipping something.

curl -fsSL https://huddle.sage-hill.workers.dev/install.sh | sh                 # start a project, get an invite link
curl -fsSL https://huddle.sage-hill.workers.dev/install.sh | sh -s --   # everyone else, in their copy of the repo

On Windows: irm https://huddle.sage-hill.workers.dev/install.ps1 | iex. No sign-up, nothing committed to your repo.

What you get

No collisions

• While an agent edits a file, everyone else’s copy goes read-only and their agent is told who has it. Agents claim the one feature they’re on (“inventory: drag and drop”) and its few files, so the rest of a level or system stays open.
• Edits land on teammates’ machines within a second or two, long before anyone commits.
• Each person’s huddle warns when their branch would conflict with a teammate’s, names the files, and one click hands the merge to an agent.

Agents that work as a team

• They message each other. “Ask dev’s agent if anyone is touching the schema” just works, and so does “tell sam I’m taking auth”.
• Describe a feature once. huddle plans it and hands each part to whoever has the most usage left in their 5-hour window. If someone runs low or goes offline, their open work moves to someone else.
• A project brief, team notes and skills that every agent reads when it starts, so nobody rediscovers the same gotcha.
• When a session ends it leaves a hand-off note, so the next agent (yours or a teammate’s) picks up where it stopped.

Work that keeps going while you’re away

• Requests sent to your agent run in the background, each in its own git worktree, up to three at once, without touching what you’re doing.
• A second agent reviews every change (from a different vendor if your team has one) and the tests run before anything merges. Failing work waits on its own branch.
• Prefer pull requests? Finished work can open one on GitHub, with the review and test results in the description.
• Schedules (“look for flaky tests every Friday”), GitHub issues sent to the team as requests, and an always-on worker for a spare machine.

Made for game jams too

• Agents work inside Unity, Godot and Blender, and read your Figma designs. Scenes and .blend files are locked like any edit, so nobody ends up with two versions of a scene that can’t be merged.