Plan, Build, and Ship — with a team of AI agents instead of one
# Add to your Claude Code skills
git clone https://github.com/openteams-lab/openteamsopenteams is an open-source multi-agent collaboration workspace. It brings multiple AI coding agents — such as Claude Code, Codex, Gemini CLI, and others — into one shared session where they can communicate, share context, and work together as a team. You can collaborate with agents through lightweight free-chat mode, or orchestrate complex tasks through structured workflows with visible plans, step-level control, and traceable review. Everything runs locally in your own workspace.
AI agents are getting stronger at planning, coding, reviewing, and testing. But more agent output does not automatically become shipped work.
Managing multiple agents is exhausting. You switch between terminals, re-explain context to every new agent, copy outputs from one prompt into the next, and reconcile conflicting diffs — your attention is drained by the chaos of juggling multiple agents.
Agent execution is invisible and uncontrollable. You command claude code to "build the feature." It runs for 15 minutes. You have no idea which subtasks it attempted, which passed, and which it silently gave up on. Most coding agents today treat a complex task as one monolithic run — there is no visible plan before execution, no way to approve or reject individual steps mid-flight, no way to retry just the step that failed. When something goes wrong, you start over.
openteams solves both problems. All agents share a single context within the same session—no more juggling between agents or repeating yourself. Complex tasks become visible, controllable workflows — you refine the plan before it runs, watch each step execute, and intervene at any node: approve, reject, retry, or redirect.
The real productivity leverage is not more agents. It is orchestrating them — with a complex plan you can see and steps you can control.
No comments yet. Be the first to share your thoughts!
You type: "Add GitHub issue sync to the workspace."
Backend: OAuth + GitHub API → Backend: Sync Engine → Frontend: Sync Status UI → Integration Tests → Final Review. Each step has a clear scope, assigned agent, and acceptance criteria.Backend: OAuth runs first. Once complete, Sync Engine and Frontend: Sync Status UI start in parallel. Every step shows its status, diff, and logs on the workflow graph.Backend: OAuth finishes. You inspect the diff, see the token refresh logic, and approve. The next steps proceed.Integration Tests fails because the sync engine returns raw timestamps instead of ISO format. You review the error log, and retry only Integration Tests step. The rest of the workflow stays intact.@Frontend Agent the sync status badge flickers when polling — debounce the state update. Fixed in one turn, no workflow needed.npx openteams-web
Download the latest release for your platform from GitHub Releases.
openteams includes a built-in openteams CLI agent. Configure your model providers in the app under menu->setting->provider config->add provider.
You can also connect supported coding agents such as:
| Agent | Example install |
| --- | --- |
| Claude Code | npm i -g @anthropic-ai/claude-code |
| Gemini CLI | npm i -g @google/gemini-cli |
| Codex | npm i -g @openai/codex |
| Qwen Code | npm i -g @qwen-code/qwen-code |
| OpenCode | npm i -g opencode-ai |
Prerequisites: Configure an API service provider or install any supported Code Agent.
step 1. Create a group chat session. Add one or more members and assign each a model and a role.
step 2. In Free Chat mode, @ any member to send a message or assign a task.
step 3. Switch to Workflow mode. Discuss requirements with the lead agent, refine the solution, and generate an execution plan.
step 4. Start the execution and review the result of each task node as it completes.
openteams supports two collaboration modes, because not every task demands the same level of structure. Think of it like Claude Code's Plan and Build modes — but for multi-agent teams: choose free collaboration when you want agents to explore and discuss openly, and structured workflows when you need reliable, predictable execution.
In free chat mode, you @ any agent to send it a task, and agents can freely pass messages to each other. Collaboration is governed by a team protocol you define — who does what, how they hand off, and what standards to follow.
free chat mode is best for small fixes, quick reviews, and exploratory discussions where a full workflow would be overkill.
Workflow mode is designed for complex tasks that need to be broken down into subtasks with observable progress and controllable execution at every step.
A lead agent drives the planning phase — clarifying requirements, designing the approach, defining the execution plan, and assigning tasks to the right agents. The result is a visible workflow with steps, dependencies, reviews, retries, and acceptance points.
Instead of asking agents to run in a loose chain, openteams turns the work into a stateful execution graph.
Note: Workflow mode uses more tokens. Please make sure your token balance is sufficient.
openteams is under active development. Here is where we are heading:
Vision: Transform token consumption into real productivity.
Have a feature reque
