Coding agents for your whole team, on your own server.

The agent setup that works for one developer breaks for a team. Zweigwerk runs Codex and Claude Code for the whole team: one login, a project set up once, and a VM per workspace that never holds your git token.

Join the waitlist

We'll get in touch by email to talk about how your team works with coding agents.

What a developer gets

A workspace is a Firecracker VM with its own kernel and its own Docker, started from a snapshot with your repositories cloned and the environment built. Run several agents at once and they all bind the same port. Agents run without permission prompts because the VM is the sandbox.

Built for a team

Your team logs in through your identity provider. An admin sets up a project once, and from then on everyone starts workspaces from it. Agents commit as Zweigwerk's git bot with the workspace owner as co-author, and admins see turns and tokens per person and project.

Part of your stack

Everything the UI does goes through the same API. A failed GitLab pipeline wakes the agent that pushed the commit, schedules start work on a cron expression, and through the API a ticket becomes a workspace, an agent and a merge request.

How the isolation works

Your git token and model login stay on the control plane, and the VM gets placeholders. An egress proxy on the host is the VM's only way out, and it swaps in the real credential on the way to the model provider, GitHub or GitLab. A VM can't reach your LAN, the host's sshd or another VM.

An agent uses both credentials through the proxy while its VM runs, so keep the git bot's memberships narrow. Project secrets are real values in the VM, so keep them low-privilege. Prompts, and the code in them, go to OpenAI or Anthropic.

Get in early

Zweigwerk is built by agents running in Zweigwerk. About 80 tasks on its own board were closed that way. It's new and it moves fast, and the teams that join early decide with us where it goes.

Leave your email address and we'll get in touch to set up a conversation about how your team works with coding agents.

For example team size, GitHub or GitLab, and how you run agents now.

We'll use these details to get in touch about early access.