Help desk
Prefill answers for the questions everyone asks. Tap a chip — no typing needed.
What is Terminus?
Guide AI
Hi! Tap a major question below, or type your own. I answer from this onboarding guide in plain English — I won’t dump assessment answer keys.
We build coding challenges that test today’s best AI coding agents. Your job: create a task that strong models often fail, but a careful human expert can still solve. Labs use those failures to see where the frontier really is.
Imagine giving an AI a laptop with a puzzle and a hidden answer key. The AI tries to solve the puzzle. Hidden tests grade pass/fail. We also keep a correct solution (oracle) to prove the puzzle is fair. If every AI always wins, the puzzle is too easy. If nobody can win, it’s unfair. We want “hard but fair.”
1) Get GitHub, Docker, Harbor, Slack, Snorkel. 2) Read this guide (especially Task + Craft). 3) Meet your track lead. 4) Put your take-home into Harbor format and open a PR within 24 hours. 5) Aim for about 5 tasks submitted in week one. Ping your lead if stuck more than 1–2 hours.
Docker is a tiny self-contained computer inside your computer. Each task runs in Docker so every AI sees the same tools and files. Install Docker Desktop, keep it running, and check `docker run hello-world` works.