Orchestration

Orchestration lets one Mel agent lead a team of agents (Mel agents, or Claude Code and Codex) that split a big task and work on it in parallel; start it with /orchestrate-agent or /orchestrate-cli. The lead agent decomposes the task, spawns workers, assigns each a scoped subtask, polls their progress, and integrates and verifies the combined result.

Two flavors

Start either one from the slash menu:

• /orchestrate-agent — the workers are Mel agents, each spawned as a full conversation in its own background session. The lead delegates rather than authoring: its own toolset deliberately drops write_file and edit_file, so the code is written by workers while the lead plans, assigns, reads results, and verifies.
• /orchestrate-cli — the workers are CLI coding agents (Claude Code or Codex) running in background panes. The lead launches each one with its full task, reads its screen to track progress, and can type into an interactive one.

The lifecycle

1. Plan — the lead thinks through the decomposition: which files each worker owns, what interfaces they share, what "done" means for each.
2. Confirm — unless autonomous mode is on (a lead starts with it on), before anything spawns a confirmation card asks "Can I start additional agents for this task?" showing the proposed fleet, which you can edit before accepting. Accept spawns; Reject keeps everything single-agent. When autonomous mode is off, Settings → Agent mode → Run orchestrated agents decides whether this card appears: Always ask (the card), Always allow (no card), or Never. With autonomous mode on, workers start without a card.
3. Spawn — workers start in background sessions that show as pills in the fleet strip above the conversation (the right / down / tab layout the lead names no longer places anything); focus stays on the lead so you can watch the coordination.
4. Assign & poll — each worker gets a self-contained task (owned files, interfaces, acceptance criteria). As they report in, the transcript shows worker status cards — an avatar monogram, the worker's name, and a "Started: …" / "Complete: …" status line.
5. Integrate & verify — the lead doesn't take worker reports on faith: it checks the actual on-disk changes (diffs, builds, tests) before summarizing, and flags discrepancies rather than hiding them.
6. Synthesize — a structured roll-up: per-worker results with what changed, anything to flag, and how to proceed.

You can also start a team from a plan: when Mel proposes a plan, Use orchestration in the plan panel hands it to a lead agent. See Planning.

Workers run hands-off

Orchestration defaults autonomous mode on, and CLI workers are launched in their own non-interactive, auto-approved modes — claude --permission-mode acceptEdits -p and codex exec --full-auto — so a nested permission prompt inside a worker can't stall the whole fleet.

Shared memory across the team

Lead and workers all get the remember / recall tools scoped to the project, so decisions made by one agent (naming, interfaces, "the API lives in src/api") are visible to the rest of the team — and to future conversations. See Project memory.

Watching it work

Every worker is a real session: click its pill in the fleet strip to open it and watch it live, or let the lead's transcript summarize. Worker conversations are saved like any other and appear in conversation history tagged by their role. Because workers edit the same checkout, git diff — and the code review panel — always shows you the team's combined output.

Run several agents yourself, without a lead

You don't need orchestration to run agents in parallel. For separate tasks you want to steer yourself:

1. Track the repository. Inside it, open the + new-tab menu and choose Track this repo for worktrees.
2. Open one worktree per task. Choose Worktree: <repo> from the same menu. Mel makes a new branch, checks it out under ~/.mel/worktrees/<repo>/<branch> and opens a tab there. See Projects & folders.
3. Start an agent in each tab. Press Ctrl+Enter (⌘↵ on a Mac) for Mel's agent, or type claude or codex for a CLI agent.
4. Watch them from one place. The Runs screen (Ctrl+Shift+M, ⌃⇧M on a Mac) lists every run under Needs you, Running and Recent, and the notification bell tells you when one finishes or waits for you.
5. Review each branch. Open code review in each tab, then merge the branches you want with git.

Use orchestration instead when the parts depend on each other and someone has to split the work and check that the pieces fit.

Limits

The number of worker agents running at once depends on your plan: 1 on Free, 3 on Pro, 8 on Max and 3 per seat on Business. See Billing & usage.

FAQ

How do I run multiple AI coding agents at once? Either start /orchestrate-agent and let a lead split the work, or open one worktree tab per task and start an agent in each.

Can workers be Claude Code or Codex? Yes, with /orchestrate-cli. They run on your own accounts with Anthropic and OpenAI.

Do workers ask me for approval? Workers run hands-off. Whether a lead may start them at all is Settings → Agent mode → Run orchestrated agents when A⁺ is off.