zeroshot By The Open Engine Company

Build your own
software factory.

CI, but it starts before the code is written.

Free and open source.

Try with your coding agent ↓Star on GitHub ↗Docs →

goalworkeracceptancereviewcode reviewboth ok?mergenorepairagainyou, reading the diffnot anymoree2e testtests the featureno double bookingon webhook retrywritesjudgesbuilt inboots the stackone topology. draw your own, optimized foryour repo, your standards, your qa process.illustrative run00:00goalno double booking on retry06:12workerdone · 4 files, 6 tests pass07:40code reviewok08:05acceptanceno · late retry books again11:31repairdone · event id kept on booking12:48acceptanceok13:02code reviewno · race: 2 at once, both book15:47repairdone · unique index, one insert16:55acceptanceok17:10code reviewok19:36e2e testno · replay gets 500, not 20021:14repairdone · replay answers 20022:20acceptanceok22:31code reviewok24:58e2e testok · 50 replays, 1 booking30:12mergeCI green · merged (--ship) goalno double bookingon webhook retryworkeracceptancereviewcodereviewboth ok?mergenorepairagainyou, readingthe diffe2e testtests the featurewritesjudgesbuilt inboots the stackone topology. draw your own, optimized foryour repo, your standards, your qa process.illustrative rungoalno double booking on retryworkerdone · 4 files, 6 tests passcode reviewokacceptanceno · late retry books againrepairdone · event id kept on bookingacceptanceokcode reviewno · race: 2 at once, both bookrepairdone · unique index, one insertacceptanceokcode reviewoke2e testno · replay gets 500, not 200repairdone · replay answers 200acceptanceokcode reviewoke2e testok · 50 replays, 1 bookingmergeCI green · merged (--ship)
Starred by engineers at
  • Google
  • Meta
  • Shopify
  • Uber
  • Booking.com
  • GitHub
  • Cloudflare
  • Atlassian
  • JetBrains

why a mechanically enforced graph?

because the agent that wrote the code shouldn't be the one that says it works.

By the time the agent says “done”, it has spent a whole session convincing itself. A skill can ask for a review, but the agent still decides whether to listen and when to stop. In a zeroshot graph, your QA process is enforced by code, and feedback is non-negotiable.

draw your own graph.

Every node is yours to change: which model runs it, what it checks, what happens when it says no.

agent that writes agent that judges built in, no model boots part or all of the stack

workeracceptancecode reviewbugfinderui revieweradversarial e2edeliverany no: back to the worker with the reason.all yes workeracceptancecode reviewbugfinderui revieweradversarial e2edeliverany no: back to the workerwith the reason.all yes

Build a review loop ↗Graph docs ↗

what a design slop reviewer catches.

The experiment: Claude Code on Opus 5.5 builds zeroshot’s landing page from its README three ways: on its own, with the reviewer’s standards in the prompt, and inside a custom graph with a design slop reviewer. To keep it a design test, that graph skips code review.

one session, no reviewer

plain Claude Code

Its final report

“I checked it with headless-Chrome screenshots at phone, tablet and desktop widths, in light and dark mode, and with reduced motion.”

open the page ↗

fly the before and after.

An arctic flight game, built with Codex. Then taken through zeroshot runs with functionality, visual and integrity reviewers. Try both.

Original build

Original low-poly arctic landscape and airplane

Open game ↗

After review + performance fixes

Later floatplane, glacial terrain and aurora

Open game ↗

WASD to fly · Shift / Ctrl for power · Touch controls on phones. One game runs at a time.

Illustrative review loop

workerbuilds the gamefunctionalityvisualintegrityall ok?approvednorepairagainyes workerbuilds the gamefunctionalityvisualintegrityall ok?approvednorepairagainyes
Run start: ribbed glacier walls and a simple red-and-white planeRun start
Candidate 1: a redesigned floatplane and instruments, the same ribbed glacier wallsCandidate 1
Candidate 2: broken shelves and floating ice replace the continuous wallsCandidate 2
Candidate 3: broader snow-capped ridges and a finished floatplaneCandidate 3
round 1round 2round 3

what users say.

Chat messages from zeroshot users. Mark Dingwall: I used Zeroshot (my fork) for my last two contracts: a production line automation and control system, and a multi-host sync and admin system for Webflow and Guesty. Very happy with it! Daniel Graungaard: I have now been running 10.7.2 for almost two days in my pipeline. Generally, I’m really happy with the results and the quality. I really like the mental models of creating profiles. It’s been so useful for me! So overall great product. Andrey: Implemented support of pi.dev harness and it works pretty well (built with zeroshot BTW). Chat messages from zeroshot users. Fav: you guys are solving this in a way that is more satisfying than the google ADK (to me) and it’s fuckin google. Trevor: you have really allowed me to just create more things in my personal time than I have before. Torstein, translated from Norwegian: it’s really nice to be able to customize the nodes and build your own loop. Jakob Thumm: I installed zeroshot and wrote my own zeroshot academic writing skill using zeroshot. It reads so much less like AI slop. I’m very happy with this!

Selected messages from our open-source adopters.

getting started.

Your coding agent can set up zeroshot, run it, and monitor progress. You approve the task before it starts.

Open Claude Code or Codex in your repo and paste this:

Install zeroshot if needed: npm install -g @the-open-engine-company/zeroshot Read the installed zeroshot skill and use it to prepare a local run for a small, non-critical issue in this repo. Stop if there are uncommitted changes. Work on a fresh branch. Show me the proposed task, how success will be checked, and the workflow. Wait for my approval before starting the run. Once approved, run it and monitor progress. Summarize what changed, what passed or failed, and what needs my attention. Suggest one useful workflow customization for this repo.

zeroshot cloud.

Run your workflow in the cloud, on your own or with a team.

Comment @zeroshot-cloud run on a GitHub issue.zeroshot Cloud ↗Cloud docs ↗

Free for one run at a time, then $100 a month for each extra run in parallel, not per seat. Models run on your own API keys.

not just
for coding.

Autoresearch runs experiments through three independent judges.

Explore autoresearch ↗
experimentevidencemethodprogressnext experiment
10 iterations. Results recorded.
Watch a research run · 20 sec

An autoresearch run

10 iterations made an exact geometry computation 21× faster on CPU.

stay in the loop.

Star it on GitHubJoin the Discord

design partners get an exclusive hoodie :)

who's building it.

Three co-founders. Eivind and Michael met doing machine learning PhD research at the Technical University of Munich. Olle has 7 years in startup growth and runs our brand.