Run it free during the beta and keep the discount.

Fresh context window on every turn, cross-LLM adversarial reviews.

In Claude Code or Codex CLI:

Create your coding plan
Run our /sprint command
Claude or Codex reviews your plan and develops the perfect coding sprint
Sprint Ox manages and orchestrates your build, and loads fresh coding agents and review agents on every turn

Format: Private beta
Cost: Free
First round: macOS

Apply for beta: limited seats.

Step 1 of 9

About 90 seconds · Free during beta · Launch discount for testers

How Sprint Ox works

An orchestrator that doesn't drift, reviews work in real time, and works non-stop.

01
You create a plan
Work with Claude or Codex to develop a coding project, just like you normally do. Greenfield or brownfield is just fine.
02
Run the /sprint command (or $sprint in Codex)
An unbiased agent reviews your plan, develops a PRD, and builds an initial sprint schedule.
03
Adversarial review
A separate agent reviews with fresh eyes, files findings by severity, and sends them back to the sprint agent to rewrite. This goes back and forth until the sprint is refined and ready.
04
Sprint Ox gets loaded and engaged
Part of the /sprint command process is asking which LLM agents you want coding and which ones you want reviewing. It'll ask a handful of other config questions, then load Sprint Ox and start the sprint loop.

Why it needs both Claude and Codex

Ask any model to review its own work and it will tell you the work is good. That is the failure mode every autonomous coding tool runs into eventually.

So Sprint Ox splits the job. One model implements. The other reviews, cold, with no memory of writing any of it. The loop only advances when the reviewer is satisfied — and the reviewer has no reason to be generous.

Do I need both Claude and Codex?

Technically no. You can run Sprint Ox on a single LLM agent. However, each model thinks differently and approaches coding differently. That fresh perspective is why these two are a match made in heaven — code with one, and review with the other.

I thought Sprint Ox was LLM agnostic?

It is, for your coding and review work. At the moment we've only developed the /sprint (or $sprint) plugins for Claude and Codex, so you start your sprint development with Claude or Codex — then you can farm off coding and review to other agents.

We plan to update the sprint creation process to work with all LLM agents in the future.

The invariant that makes it safe: control flow is deterministic code, not a model decision. AI is trusted for judgment about quality — never for deciding what runs next, when to stop, or what counts as done.

The deal, both directions

A beta is a trade. Here is exactly what each side is putting in.

What you get

Free access for the whole beta — no card, no trial clock.
A discount that carries over when it ships as a paid product.
A direct line to me for bugs, not a support queue.
Early say in what gets built next, while it's still cheap to change.

What I need from you

→Run it on something real — a project you actually care about finishing.
→Tell me when it breaks, with enough detail that I can reproduce it.
→Tell me when it's confusing, which is more useful and harder to get.
→A couple of honest hours in the first two weeks. That's the whole ask.

Apply to test Sprint Ox →

Questions before you apply

What is Sprint Ox, in one sentence?+

It's an autonomous sprint orchestrator that runs on your own machine: you hand it a backlog, Claude Code implements each sprint, Codex adversarially reviews the work, and a deterministic loop repeats test → review → fix until every critical, high, and medium finding is gone before it moves to the next sprint.

Do I really need both a Claude and a Codex subscription?+

Yes. The whole point is that the model writing the code is never the model judging it. Claude implements, Codex reviews with fresh eyes, and neither one gets to mark its own homework. With only one of them you can run half the loop, which is not the thing being tested.

Does it push to my repo or deploy anything?+

No. It works on per-sprint git branches with checkpoint commits, and those commits stay local — the loop never pushes. Nothing leaves your machine unless you push it yourself.

What happens when I hit a usage limit mid-sprint?+

It pauses and resumes. Hitting a limit is expected on long runs, so the loop treats it as a pause rather than a failure and picks up where it left off. It also rotates context at around 70% by writing a handoff document, so long sprints don't degrade as the window fills.

Is the beta free? What happens when it launches?+

The beta is free. Testers who actually run it and send feedback keep a discount when it ships as a paid product. No credit card, no trial timer.

What if I'm on Windows or Linux?+

Apply anyway, but be honest on the hardware question. Today's build ships a macOS background service and desktop dashboard, so the first round of testers will be on Macs. Everyone else stays on the list for when that changes.

How long does applying take?+

About 90 seconds. Two fields, then eight one-tap questions and one short written answer. The written answer is the one that matters most.

Start here

OpenClaw + OpenAI Codex ChatGPT Setup

If your bots are breaking during the Claude-to-Codex transition, this is the first page to read. It covers the install walkthrough, the migration prompt, and the safer one-bot-at-a-time path.

Open the guide →