Orchestration
Orchestrate a fleet of coding agents
You’re the operator. Decompose a goal into a plan, launch agents in parallel — each isolated in its own worktree — and watch every session from one console that tells you exactly who needs you next.
Each agent, its own worktree
Every agent works in a dedicated git worktree — no merge conflicts, no stepping on each other’s changes. Diff, review, and merge each branch back only when you say so; until then, nothing touches your working tree.
Five verified adapters
Claude Code, Codex, Cursor, Gemini CLI and Aider, each on persistent daemon-backed terminals with split panes that survive restarts and view-switches. One click opens anything an agent touched in VS Code, Cursor, Zed, or JetBrains. Additional experimental adapters sit behind a toggle and are not supported for production use.
Every session on one live grid
A needs-attention queue puts what’s waiting on you first — awaiting-input, blocked, and errored sessions, oldest waiter at the top. A global kanban rolls up every agent’s tasks and live status so you always know who’s working and who’s stuck.
A goal becomes a plan
Hand Operon a goal and it decomposes it into a subtask DAG, each subtask carrying a tool recommendation and its own session link. Subtasks run one at a time in dependency order — watch the plan advance as each session completes its piece before the next starts. (Concurrent subtask execution is on the roadmap; Fleet and Arena already run independent sessions in parallel today.)
Agents talk to each other
Agents pass messages, share a task DAG, and open ask-human gates when they need a decision — and those gates surface directly in Mission Control so nothing blocks silently in a terminal you’re not watching.
FAQ
Orchestration, answered
Q
Operon does not impose a fixed limit.— practical concurrency depends on your machine and on each agent’s own rate limits. Operon puts every session on one live Mission Control grid, so whether you are driving two agents or a dozen, you see each one’s status, cost, and activity in a single view — with a needs-attention queue that surfaces exactly the ones waiting on you.
Q
Yes — its own isolated Git worktree.Every session works on a separate checkout, so agents never collide or step on each other’s changes. When a session finishes, its worktree is reclaimed automatically, and any uncommitted work is preserved rather than destroyed.
Q
Claude Code, Codex, Cursor, Gemini CLI, Aider, and more.Operon drives the five core tools with dedicated adapters plus a broader catalog of CLI agents through a generic launcher — so you can run a heterogeneous fleet from one operator console.
Q
Through a built-in coordination bus.Agents can send each other messages, claim tasks from a shared DAG, and raise decision gates — including "ask a human" gates that block a run until you answer. It is inter-agent messaging without you having to relay context by hand.
Q
No.The in-app Git Workbench takes you from status to diff review, staging, commit, push, and pull request without ever dropping to a terminal — and the winning run from an Arena race routes straight into that same merge flow.
Get started
One operator, many agents.
Private beta — join the waitlist to get an invite when your slot opens up.
Related reading
- Run multiple AI coding agents in parallelIsolated Git worktrees and one Mission Control grid for every session.
- Compare AI coding agents on the same taskRace two to four agents on one goal and judge them on cost and diff quality.
- Multi-agent developmentVolume VI — when coordinated agents help, and when they cost more than they return.