Skip to main content

See how SpecWithCode runs your agents

You know how machines, subscriptions, usage limits, the coordinator and its threads fit together before you install.

  • SpecWithCode
  • 2:54
  • Claude Code and Codex
  • Narrated, with captions

Recorded on SpecWithCode 1.0.5, added 13 October 2026. The current release is SpecWithCode 1.0.44.

Length 2:54Captions

Steps

  1. Sign each machine in with the official tools. Keep every machine signed in to one account per provider for good, for example Claude Code on one account and Codex on another, and add machines when you need more capacity instead of switching subscriptions by hand.
  2. If you have one subscription, run it on the machine with the most free memory and disk, because every session runs on that machine.
  3. Connect your machines privately. Each machine runs its own SpecWithCode server.
  4. Watch the limits: SpecWithCode shows session and weekly usage, banked resets and credits for every account. When one runs out, a thread waits for the reset or moves to another provider or machine.
  5. In a project, talk to one coordinator. It plans, then starts threads for the work, each on the machine and model that fits, and each thread opens its own pull request.
  6. Open a thread to see what it ran and what it is waiting on. A thread resolves itself only when its pull requests are merged, checks pass and the review is approved.

Transcript

Built from the caption file of the narration.

0:00 More AI than you can use

You probably pay for more AI than you can use. Claude on one machine, Codex on another, maybe Grok or Copilot too. SpecWithCode turns all of it into one team.

0:13 One machine, many subscriptions

Start with the basics. Every machine signs in with the official tools, and one machine can hold several. My M4 Max runs Claude and Codex on the EasyChamp admin accounts. The M1 runs Claude and Codex on the marketing accounts. Nothing is resold.

0:31 Only one subscription? Pick its host

Only have one subscription? Then choose the machine that hosts it with enough memory and free disk space because every session runs right there on your machine, not in someone else's cloud.

0:44 Connect your machines

Every machine runs its own SpecWithCode server. You connect them privately over your network, an SSH tunnel, or Tailscale. Here, the M4 Max sees the M1 and the M3, and each one is paired on purpose.

0:59 Limits, CLIProxyAPI and fallbacks

Then, SpecWithCode watches the limits. Session and weekly usage, banked resets, and credits for every account. Pool accounts behind a CLIProxyAPI hub, and they show up too. When one runs dry, a thread waits for the reset or moves to another provider or machine.

1:18 One coordinator, many threads

Now the part that changes how you work. In a project, you talk to one coordinator. It plans, then starts threads for the work, each on the machine and model that fits.

1:30 A real project: EasyChamp

This is EasyChamp, a real project. One thread is shipping reusable tactical templates. Another built club finance and is waiting for my review. Others ran independent reviews. Each one opens its own pull request.

1:44 Inside a thread

Open a thread to see exactly what it ran and what it is waiting on. Reply and your message goes through the coordinator with the thread as context.

1:54 Verified, and auto-resolve

And done is not the same as verified. You can check the work yourself or let auto resolve do it. A thread closes itself only when every pull request is merged, checks pass, and the review is approved. Anything uncertain waits for you.

2:11 Which machines: Linux, Windows, Mac

Which machines? Linux makes the fastest workers, and Windows is a solid worker too. A Mac works, but APFS is slow for agent workloads. So let the Mac orchestrate. Chrome and marketing tools, iOS builds, and computer use.

2:28 A good worker machine

For a worker, aim for eight cores, 32 gigabytes of memory, and a fast SSD with a few hundred gigabytes free. Budget about four gigabytes of memory and 10 of disk for every agent running in parallel.

2:42 Get it

Your machines, your subscriptions, one control desk. SpecWithCode, open source at spec-weave.com/code.

Download SpecWithCode for MacRead the reference

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 (this guide)

    2:54, SpecWithCode

  2. Run a project with a coordinator and workers

    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