Tutorial: Polly (Multi-AI Coding)

Polly is a multi-agent coding orchestrator. You give it a task; Polly splits it into sub-tasks and uses intelligent routing to pick the right agent for each one (Claude Code, Codex, Cursor, and others). Then a different agent reviews the results: if Claude Code writes the code, Codex checks it. Each sub-agent gets a clean slate with no prior conversation history, so it stays focused.

Time: 5 minutes. Everything runs locally.

1. Start Polly

Launch Polly. Either command starts it in your terminal and opens a web UI at http://localhost:6767.
```
omni polly    # explicit
omni          # same, when a Claude credential is configured
```
Start Polly with omni polly, or just omni.
Give Polly a task, for example:
```
Refactor the authentication module into separate files for OAuth, JWT, and session handling. Add tests for each.
```
Polly breaks this into sub-tasks and delegates each one to a different AI agent. Claude Code implements one piece, Codex implements another in parallel, and a different agent reviews each PR.

2. Watch the agents work

Claude Code and Codex work on different parts of the task at the same time.

Open http://localhost:6767 or launch the desktop app and connect to the same address.

Click the Agents tab to see every sub-agent Polly dispatches, with live status (working, idle, finished).

Click any sub-agent to switch to its view and see its conversation, files, and terminal output.
Intervene at any point: send a message to a sub-agent, or switch back to Polly to see the orchestrator's view.
Switch to the Shells tab to see all running shells side by side.

Switching from Polly to sub-agents in the Agents panel — The Agents panel shows Polly and its sub-agents, so you can inspect or steer each delegated worker directly.

3. Turn on Smart Routing

Smart Routing picks the best model for each sub-task based on complexity: a short, well-scoped task goes to a faster model; an ambiguous or cross-cutting refactor goes to a stronger one. Polly applies this per sub-agent spawn, so each worker gets the model suited to its slice of the work.

To enable it, click the gear icon in the chat composer and select Smart Routing from the Model dropdown. A chip appears below each routed message showing which model was picked and why.

Smart Routing requires a Databricks AI Gateway connection. If you're running entirely locally, the router is unavailable and Polly falls back to the default model.

4. Use Polly's built-in skills

Type any of these into the Polly chat to go further:

/fanout: run independent tasks in parallel. Each gets its own sub-agent and git worktree, and opens its own PR.
/cross-review: send an implementer's diff to a different-vendor reviewer. Blocking issues loop back as fixes until clean.
/investigate: delegate read-only work (debugging, audits, code understanding) to sub-agents and get a synthesized report.

See the Polly reference for full details.

What's next

Collaborate from anywhere to deploy a server and share sessions with teammates.
Build a custom agent with your own multi-agent orchestration in YAML.