Build your own software factory of coding agents

bb’s thread list marks every agent as running, waiting on you, or done, across all your machines. When one needs an answer or an approval, the bb mobile app sends a push notification, and you can reply from any browser.

Yes. Give each thread its own Git worktree so agents don’t overwrite each other’s changes. List your .env files and setup commands once, and bb prepares every new worktree. By default, bb runs one thread per processor core and starts the rest as others finish.

Yes, and there’s no coordinator to set up. Tell your agents to work together and they coordinate on their own: any agent can message another thread, whatever the provider, agents answer each other’s messages, and an agent hears back automatically when a thread it started finishes or fails. The receiving thread shows each message and who sent it, and you can step in and message any of them yourself.

Yes. Ask Claude Code to “start a bb Codex thread to review this branch, then fix what it finds.” It starts Codex in the same worktree, waits, and applies the fixes. Codex shows up as its own thread, so you can read the exact prompt Claude sent, watch it work, and message it mid-run. Any other pair works the same way, like Cursor and OpenCode.

Yes. bb works out of the box, with worktrees, diff review, automations, and the mobile app ready from your first thread. When you want more, install plugins from the marketplace or ask an agent to build one. Plugins can add panels, commands, and new agents, and they work in the mobile app too.

Yes. The bb CLI covers everything the app does: start and message threads, run terminals, schedule automations, and manage machines and plugins, so a script or another agent can drive bb end to end. The SDK and plugin API let you add your own commands, tools, and screens.

bb picks the work back up. When an agent stops on a usage limit that reports when it resets, bb sends the message again a little after the reset, up to four times per turn, so you don’t have to come back and press send. Credit and spend limits aren’t retried.

Yes. Your agents run on your own machines, so the computer running bb has to stay awake and online. Switch on Keep Awake, a built-in bb plugin, to stop idle sleep on macOS and Windows. To step away with your laptop closed, run bb on an always-on desktop, home server, or cloud VM.

Claude Code, Codex, Cursor, and OpenCode, plus Pi, Grok Build, omp, and Hermes Agent. In Settings, add any other agent that supports the Agent Client Protocol (ACP), an open standard for connecting coding agents to apps, like Gemini CLI or Devin. You can also add one with a plugin. Each thread can use a different agent.

1. Download bb for macOS (Apple Silicon), Windows, or Linux (both alpha). On an Intel Mac, run npx bb-app@latest, which needs Node.js 22.19 or later.
2. Install and sign in to a coding agent, like Claude Code, Codex, Cursor, or OpenCode.
3. Open bb, add your project folder, and type your first request.

Yes. bb runs the agents you already use, signed in the way you already pay for them: a Claude Pro or Max plan, a ChatGPT plan for Codex, a Cursor plan, API keys, or any other provider’s plan. Your existing setup comes along too (CLAUDE.md, skills, MCP servers, and agent settings).

Yes, for one person or a whole team. The mobile app, remote machines, automations, and plugins are all included. You pay only for the agent plans or API keys you already use, from any provider. The code is on GitHub under the MIT license.

Yes: bb runs on your machines, so your code goes only where your agents send it, the AI providers you chose. Anonymous usage stats exclude code and prompts and can be turned off. The bb Connect relay doesn’t store traffic, but isn’t end-to-end encrypted.