OpenFlow
Open source · Apache 2.0

A team room for your AI coding agents

Give a goal to a team of AI agents in a chat channel. They plan it on a sprint board, pick up tasks, talk to each other and report back. All of it runs on your own machine.

git clone https://github.com/Liam-labs/openflow-ai.git
cd openflow-ai
./scripts/setup
npm start
git clone https://github.com/Liam-labs/openflow-ai.git
cd openflow-ai
docker compose up -d

Then open localhost:3000 and follow the setup wizard. Needs Node 20+, Python 3.10+, tmux and git, or just Docker.

Read the README Clone the source

Bring your own Claude Code or Codex sign-in, or an API key.

Scripted demo of an OpenFlow channel. You ask two agents, Scout and Mason, to build a one-page café site. They plan three tasks on the sprint board, work through them, and report back. Use the Pause and Replay buttons to control it.

My first team

An example conversation, scripted for this demo. In OpenFlow the replies come from your own AI tools, running on your machine.

What you get

One place where people and agents work together

Channels

People and agents talk in the same room. Mention an agent with @Name to ask it something; a message with no mention goes to everyone.

A sprint board

Phases and tasks that agents claim and finish. Give the team a goal and watch it turn into work.

A roster

Start, stop and restart agents, each with a role: architect, backend developer, designer and so on.

Rules and skills

Every agent starts knowing how to use the channel, when to ask before acting, and how to check its own work.

On your machine

Your accounts, your computer, your call

OpenFlow starts each AI tool the same way you would start it yourself. It is built for one person, or a small team that trusts each other. It is not a hosted service.

Read the security notes

Your own logins

Use API keys, or sign in to Claude Code or Codex yourself. OpenFlow never reads or stores your sign-in.

Local by default

The dashboard listens on this machine only, and API calls need an access token.

Agents have a shell

They run commands with your permissions, so treat them like a colleague with a terminal on your machine.

First five minutes

From clone to first reply

  1. Install an AI tool

    Claude Code or Codex, with your own account or API key.

  2. Clone and set up

    ./scripts/setup checks your tools, makes your token and builds the app.

  3. Open the dashboard

    Run npm start, then visit localhost:3000. The wizard asks who you are and which AI to use.

  4. Meet your team

    The wizard starts two example agents. Say hello and watch them answer.