CARLOS rastrillo docs

๐Ÿค– Compare

Rastrillo is designed specifically for CARLOS: built with AI in mind (but AI not required), to be deployed with high-resilience and low cost.

Everything validated against each project's repo 2026-08-28.

Routers

Simple frameworks for routing and middleware. All four MIT licensed.

Latest Direct deps Handlers are
chi v5.3.2 none http.Handler
echo v5.3.1 3 echo.Context
gin v1.12.0 15 gin.Context
fiber v3.5.0 14 fiber.Ctx

Rastrillo's middleware is func(http.Handler) http.Handler and its handlers take (w, r), so chi drops straight in. The other three need adapters. Scaffolded apps use chi for its route groups, though the framework itself imports no router at all.

Fiber is built on fasthttp instead of net/http. That buys speed. stdlib middleware no longer usable.

Full-stack frameworks

Latest Licence Frontend Data Deploys to
Buffalo v1.1.4 MIT Plush templates Pop anywhere
Encore v1.58.4 MPL-2.0 none โ€” API-first own SQL primitives Encore Cloud, your AWS/GCP, Docker
Goa v3.30.0 MIT none โ€” generates clients none anywhere
Velocity v0.76.2 MIT Inertia + React/Vue + Vite own generic ORM anywhere
Rastrillo v0.19.0 MPL-2.0 html/template + ui GORM + SQLite CARLOS

Encore needs its own CLI and compiler. Everything else here builds with go build. Velocity is pre-1.0 and moving fast enough that it has already retracted a v1.0.0 published by mistake.

Where the others win

Reach for a router โ€” chi, echo, gin, fiber โ€” when you are building a JSON API and want to pick your own database, auth and deployment. A router plus four libraries you chose is a legitimate architecture, and the most common one in Go.

Reach for Encore when you want infrastructure to follow from the code, across Postgres, Pub/Sub and cron, provisioned into your own cloud account. Rastrillo is a different direction.

Goa suits a project where the API contract is the artifact: define it once, and OpenAPI, gRPC and typed clients all generate from that design.

Velocity is the closest thing here to a full-stack framework with a modern frontend attached โ€” thirty-odd services under one API, React or Vue over Inertia. It is much younger than the rest. Buffalo has the same ambition and a decade of history behind it, still maintained, moving slowly.

Where Rastrillo is the wrong answer

You need CARLOS โ€” see Deploying. CARLOS prefers server-rendered apps with progressive enhancement. If you want a React or Vue frontend there is no Inertia adapter and no asset pipeline. Assets keeps things very simple: no bundling, just fingerprinting.

Velocity has more JS tooling. Rastrillo assumes SQLite, so Postgres and MySQL have to be own-rolled โ€” Data has the detail. And for a pure JSON API, sessions, CSRF, flash and owner-scoped 404s are most of what you would be paying for.

The last one is the real filter. Since Rastrillo is built by AI for AI, the SKILL.md budget that shapes so much of the design is irrelevant to you.

What the budget buys

The other frameworks here optimise for what a developer can build quickly. Rastrillo optimises for what an agent can hold at once: one 15 KB SKILL.md, budgeted and enforced by a test, covering the whole framework. Rastrillo aims for token efficiency and thus time and cost efficiency.

Nothing else on this page ships a single-file spec an agent loads before it starts. In an AI-first world, Rastrillo is an AI-built solution.

You can still write every line yourself; the budget just stops being the reason to choose it.

Whether that matters depends entirely on who is writing the code.

Read this page as markdown โ€” exact, unstyled, and cheap for an agent to fetch.