Skip to main content

Run a project with a coordinator and workers

You can tell the coordinator and worker models apart, follow the routing history, resolve and reopen a thread, and read token usage by model.

  • SpecWithCode
  • 1:29
  • Claude Code and Codex
  • Captioned, no sound

Recorded on SpecWithCode 0.2.0, added 9 October 2026. The current release is SpecWithCode 1.0.44.

Before you start: SpecWithCode installed; Claude Code and Codex signed in.

Length 1:29Captions

Steps

  1. Set the coordinator and worker defaults separately. Existing threads keep their own model choices.
  2. Talk to the coordinator. Its conversation stays in the project and keeps the routing history.
  3. Open a worker thread on the right to see its details while the coordinator stays in view.
  4. Choose Resolve when work is finished. History is kept, and Reopen brings the same thread back when related work returns.
  5. Open Usage to compare coordinator and worker token totals by model. Cache reporting can be incomplete, and an API-equivalent cost is not money paid.

Transcript

Built from the on-screen captions of this silent video.

0:00 What this run verifies

Verified here: native project coordination. Device deployment is checked separately.

0:09 Coordinator and worker models

Coordinator and worker defaults are independent. Existing threads retain their own model choices.

0:18 The Codex coordinator

This Codex coordinator completed the verified native run. Its conversation stays in the project.

0:30 A Claude worker thread

This Claude conversation completed in the verified native run. No new inference starts in this recording.

0:43 Routing history and worker details

The coordinator keeps the routing history while worker details open on the right.

0:53 Resolve and reopen a thread

Resolve organizes finished work without deleting its history. Reopen the same thread when related work returns.

1:09 Usage by model

Open Usage to compare the actual coordinator and worker token totals by model. Thread shares explain where tokens went. Cache reporting can be incomplete, and API-equivalent cost is not money paid.

Read the projects guide

Learning path

Run SpecWithCode

Understand the setup, run a project through its coordinator, then add plans, memory and a second provider.

  1. See how SpecWithCode runs your agents

    2:54, SpecWithCode

  2. Run a project with a coordinator and workers (this guide)

    1:29, SpecWithCode

  3. Plan, verify and schedule project work

    1:35, SpecWithCode

  4. Keep project memory and read honest usage

    1:06, SpecWithCode

  5. Continue a thread on Codex at a usage limit

    1:01, SpecWithCode

5 more guides are planned for this path.

All tutorials