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

Rust off Rails is an opinionated web stack for the AI era, written in Markdown files that guide coding agents through the process of building a full-stack Rust web application. The framework emphasizes a single, uniform approach to each aspect of development, from the stack itself to the layout and the way to handle various tasks.

The agent generates code based on a set of product questions, each with a recommended default. Installation requires running an install script and providing answers to product questions. The framework provides a compact plan with resources, pages, jobs, and assumptions, along with acceptance criteria. The spec serves as the durable record, while chat is not.

The generator recipes, run by the agent, follow a specific sequence: migration, model, failing tests, controller, and views. Each step is executed one vertical slice at a time. The code includes formatting, Clippy checks, tests on SQLite and Postgres, and coverage gates. The goal is to have green status on both databases before the work is considered complete.

The dashboard is lightweight, clocking in at about 1 µs per request in a release build and is only used during development. The technology stack includes Axum, Tokio, SeaORM, SQLite, Postgres, Askama, Tailwind, lettre, tracing, and one binary with a single Docker image. Askama templates are checked at compile time, and htmx is used for added functionality. The framework supports CRUD apps, dashboards, content sites, and React or Svelte pages with server-side routing and controllers through a generated Inertia adapter.

React or Svelte pages are used for standalone apps requiring a JavaScript UI. A separate single-page app with a JSON API is available for standalone apps. Types flow from Rust using ts-rs, and the framework can be used to build a backend with a versioned API for mobile apps, partners, and scripts.

The database is the only infrastructure required, featuring a job queue, cache, and pub/sub capabilities. The database is built with GPT-6 Luna Claude Sonnet 5.5, Meta's Muse, Wonder 1, GLM 5.3, Flash, and via opencode. The ultimate goal is to create small, cheap agents capable of delivering real work, with documentation available on GitHub.