Built for AI agents
The Rails template built for AI coding agents.
Most templates were written for humans and retrofitted with a README. Uno was structured from the start so Claude Code, Cursor, and any capable agent can build production features in it reliably — and cheaply.
What is an AI-native Rails template?
An AI-native template is a codebase a coding agent can navigate, extend, and verify without a human explaining it. Concretely that means machine-readable conventions (AGENTS.md), retrievable per-feature documentation, views in plain code rather than template dialects, hard boundaries between domains, and a test suite strict enough to catch generated mistakes. Uno is a one-time-purchase Rails 8 SaaS template built exactly this way.
Six things that make it agent-native
AGENTS.md is the contract
One file at the repo root defines the rules: routing, conventions, the doc map, what an agent must read before touching a layer. Claude Code, Cursor, and any agent that reads instructions starts from the same paved road instead of guessing at 2am.
A doc per feature — 40+ of them
Auth, billing, admin, jobs, notifications, event sourcing, encryption, deploys: each feature ships a doc that says what it does, where it lives, and how to extend it. Agents retrieve the one doc they need and ship in a single pass — no repo archaeology.
Phlex views: pure Ruby, fewer tokens
Views are plain Ruby objects. No template dialect to mis-edit, refactorable with the same tools as the rest of the code, and far cheaper in tokens than ERB/HTML — which is what your agent bill and context window are made of.
A test suite that reviews the agent
Integration tests, factories, and an N+1 detector that fails the build. When the agent ships slop, the suite catches it before production does. bin/lint and bin/rails test are the review gate the agent can run itself.
Domain contexts the agent can't blur
Features are isolated in contexts under app/core with documented boundaries. The agent extends one context without leaking into the next — feature #50 lands as cleanly as feature #5.
AI infrastructure already inside
Multi-provider LLM clients, usage logging, a per-call audit trail, and a tool-calling agent framework. The product you build on Uno can ship its own AI features without new plumbing.
What the workflow looks like
You describe the feature. The agent reads AGENTS.md, pulls the relevant feature doc, and follows the documented conventions for the layer it touches — models in domain contexts, views in Phlex, jobs on Solid Queue. It runs bin/lint and the test suite before calling anything done; the N+1 guard fails the build if it regressed a query. Your job shrinks to what only you can do: deciding what to build and reviewing what shipped.
Why do tokens matter?
Agents pay per token, twice: once in money, once in context-window attention. ERB templates mix HTML, Ruby, and dialect syntax the agent must parse and reproduce exactly. Phlex views are Ruby all the way down — shorter to read, safer to edit, cheaper to regenerate. Across hundreds of agent edits, the difference compounds.
When this is the wrong page
If you don't use AI agents at all, Uno is still a solid Rails 8 template — but this page's argument won't move you; read the stack rationale instead. If you want an agent to generate your entire app from scratch, no template will save you from reviewing auth, billing, and encryption by hand — that's precisely the work Uno ships pre-verified. And if your product is team-shaped B2B with deep multitenancy, look at Bullet Train's team model before deciding.
Point your agent at something finished
One-time purchase. The boring 90% already wired, documented, and tested.