wayformFAQGet early access What is agent orchestration for teams?
Agent orchestration means deciding what each coding agent works on, keeping its work apart from everyone else’s, and putting a person in front of the decisions that matter. Wayform does this from a macOS app. Each plan builds in its own branch with the agent you choose, briefed by your team’s decisions, and a failing pull request can wake its build to fix it.
Your agents run on your machine with your own subscriptions. Wayform doesn’t host a fleet of cloud agents; it coordinates the ones you already use.
How it works in Wayform
- Start from a plan. Open an agreed plan and choose Build. Pick Claude Code, Cursor, or Codex, plus a model and reasoning effort.
- One branch per build. Each build runs in its own worktree on a
wayform/<plan>-<agent>branch, so parallel builds and your working copy don’t collide. - Each task is briefed. Subagents get the decisions that bind their checklist item. The build chat shows progress against the plan’s own checklist.
- Approve, or run on Auto. Actions that need a person arrive as approval cards, one at a time, in the dock where you’d type. With Auto on (Claude Code and Cursor), the agent’s own reviewer handles routine actions. Wayform’s deny rules apply first, and risky or irreversible commands still come to you.
- Steer as it runs. Send follow-ups mid-build. Check usage with
/usage. - Check, then open the PR. The agent grades its diff against the plan before the pull request.
- Wake on red CI. If the linked PR’s checks fail, Wayform resumes that build with restricted access to read the failure and push a repair (automatic for Claude Code and Cursor). Wakeup turns never run in Auto mode.
An example
Two plans are active: #12 team invitations and #14 a usage dashboard. Maya builds #12 with Claude Code on Auto; Chris builds #14 with Codex and approves each step. Each build has its own branch. When #12’s PR fails a test, its build wakes, fixes the test, and pushes. Maya reviews a green PR in the morning.
What it is not
- Not a cloud agent platform. Builds run locally, using your agents and your accounts.
- Not a security sandbox. Auto mode’s reviewer and Wayform’s rules reduce risk; they are not a security boundary.
When you don’t need it
If you run one agent at a time in your own terminal and review every step, the CLI build is enough. Orchestration helps when several builds run at once, or when people step away while agents work.
Go further
- The desktop app: builds, Auto, and wakeups in detail.
- Guardrails for coding agents.
- Shared plans.
Questions
Can I run several coding agents at once?
Yes. Each build runs in its own worktree on its own wayform/<plan>-<agent> branch, so parallel builds don’t collide with each other or with your working copy.
Does Wayform run agents in the cloud?
No. Builds run on your Mac with your own Claude Code, Cursor, or Codex accounts. Wayform coordinates them and keeps the shared plans and memory.
Which agents support Auto mode?
Claude Code and Cursor. In Auto, the agent’s own reviewer handles routine actions, Wayform’s deny rules still apply first, and risky commands still come to you. Codex builds use approval cards.
What happens when CI fails on the pull request?
Wayform wakes the build that opened it. The agent resumes the same chat with restricted access, reads the failure, and pushes a repair. This is automatic for Claude Code and Cursor, and wakeups never run in Auto mode.