Libraries

Norma ships with the compiler and you can use it immediately. Triga, Gradus, and Tela are optional dependencies you declare when you need graphics work, automatic differentiation, or typed views. Inferentia is an application built on them, still in development.

Norma — bundled#

The standard library. Imported as norma:*, no installation step, no entry in faber.toml:

import from "norma:solum" only

import from "norma:textus" string

main {
    print "ready"
}

It covers I/O, collections, strings, math, and system interaction. It is deliberately small — the surface that exists is stable, rather than broad and provisional. Norma →

Triga — opt in#

A native graphics and geometry engine written in Faber: math, scene graph, materials, geometry, and resource types as triga:* modules. Declare it as a Cista dependency in faber.toml and the compiler resolves it from the package store.

Its shapes model the three.js abstractions so that models and people already familiar with that vocabulary can move quickly — but Triga is not a binding to three.js. These are native Faber types compiled to GPU shaders. The frames on the home page were produced by it. Triga →

Gradus — opt in#

Automatic differentiation, losses, optimizers, and neural-network primitives as gradus:* modules. Reverse-mode gradients are a Radix compiler transform — a function annotated @ radix backward gets a generated companion that computes its gradients — and Gradus is the library surface over that mechanism, so model code calls functions instead of hand-writing annotations.

It is JAX-shaped: models are pure (params, x) → y functions with explicit parameters, and there is no module class hierarchy or runtime tape. It is also pre-1.0 and narrower than its module list suggests — shapes are concrete overloads rather than generics, and training is CPU-correct rather than fast. Gradus →

Tela — opt in#

A view library: typed HTML and SVG values, a fail-closed serializer that turns a valid tree into markup, and the DOM and Canvas2D contracts for a browser host. Imported as tela:*. It is early (version 0.0.0): the view protocol and static renderer work, while its Rust output and parts of its own proof suite do not yet. Tela →

Inferentia — an application#

A local-first inference server written in Faber: a command line and a small HTTP API over a GGUF model, built on Gradus and Norma. It is not a library and not a released product. The page states what exists and what does not. Inferentia →

The corpus — not a library#

The corpus is the language's worked-example tree: one directory per construct, each holding small programs that exercise it. It is the source of truth for what the compiler supports, it generates the target matrix, and it is browsable at /corpus/.

You do not import it. You read it when you want to know how a construct actually behaves rather than how it is described. About the corpus →

Writing your own#

A library is a package with kind = "lib" in its faber.toml and a [library] provider name. faber install puts it in the Cista store; other packages resolve it through faber.lock. See Packages with Cista.