ν Nomos
Architecture

How Nomos works, end to end.

A .nomos file becomes a verdict with a proof tree in four stages. Everything is deterministic except the fact resolver — which is why the fact resolver carries confidence, authority, and a path to human review.

Every box below links to the actual file in the repository. Nomos is ~15% new code; the rest stands on the shoulders of permissively-licensed OSS — see the stack at the bottom of this page.

§ 01

The compiler pipeline.

input

The user writes Nomos source + supplies facts via JSON or by pointing at source documents the LLM will read.

transform

Lex → Parse → Check → Resolve facts (via LLM if any). The fact resolver is the only non-deterministic stage.

output

A Value + a proof tree. The proof tree names every rule, requirement, authority, and fact used — auditable end to end.

§ 02

Surfaces.

The same core compiles to multiple surfaces. One parser, one evaluator, one source of truth — used from a terminal, a browser, a proxy, an editor, or an agent.

§ 03

What we stand on.

Nomos is not built from scratch. The field has fifty years of work; we extend it. Every tool below is permissively licensed and battle-tested somewhere that matters.

Next

The thesis →

Why Nomos exists, and what makes its wedge defensible after fifty years of rules-as-code attempts.

Context

Prior art →

Catala, OpenFisca, Blawx, Kowalski's British Nationality Act, and the others. What we borrow, what we leave.

Build log

Changelog →

Every slice, shipped in public. Nomos is built in the open, commit by commit.