OpenAnvil docs
Understand the system before operating it
Start with one complete run. Then explore mobile grading or the cloud platform without mixing runtime behavior, deployment, and future design.
Start here
Architecture
→Follow one run from trial loading through the agent loop, grading, reward, and artifact publication.
Deep dive
Mobile grading
→See what boots Android and iOS, which tests run, and how deterministic reward is calculated.
Deep dive
Hosting and deployment
→Understand private app hosting, the event control plane, S3, compute, and the three deployment workflows.
The short version
Give an AI a trial
Specification plus visible contract
→
Let it build the app
Controlled tools and full trajectory
→
Grade with hidden tests
Evidence plus deterministic reward
How to read implementation status
Working
Implemented behavior with a usable code path.
Partial
Implemented pieces that still need end-to-end wiring or validation.
Target
Intended architecture that must not be described as shipped.