Rust off Rails: The opinionated web stack for the AI era

Rust off Rails is an opinionated full‑stack web approach that ships not as a library but as 60 Markdown decision files that instruct coding agents how to generate complete Rust web applications. Humans supply product intent through a guided interview or plain English specs, and agents perform an interview → plan → generate → green workflow that yields migrations, models, controllers, frontend pages, tests and a deployable binary. Every line of code - Rust, SQL, HTML, CSS, TypeScript, Dockerfiles - is produced by the agent; conventions (crate choice, file layout, naming, tests) are fixed in prose so the agent never debates choices. The stack is one crate per job using Axum, Tokio, SeaORM, Askama, Tailwind, lettre, tracing and a single Docker image; frontend modes include htmx (default), inertia with React or Svelte, SPA or API. The database is treated as the sole infrastructure providing queue, cache and pub/sub.

Practically, agents run generators that create failing tests and iterate until fmt, clippy and test suites pass on both SQLite (local) and Postgres (prod), tracing each acceptance criterion to a test. Built‑in developer tooling includes request traces, a hot‑functions profiler, a dev mailbox, jobs dashboard and presence. Commands like /offrails:new, /offrails:feature and /offrails:deploy manage lifecycle tasks. Field tests show agents porting real apps with green test runs. The core argument: when typing is free, explicit, compiler‑checked code and rigid conventions let agents reliably produce correct, auditable applications.