Get started

Introduction

Nullius runs AI evaluations on your machines and opens every score into the evidence behind it.

Do	How
Run	A benchmark against agents, capabilities or its own setup, on the Nullius standard or Harbor
Open	Nullius and Harbor jobs; ATIF and OpenTelemetry exports, converted on open
Inspect	Trials, conversations, traces, verifier reports, judgements; no model key
Compare	Runs on their shared tasks
Bring in	A benchmark, agent or capability, by one prompt or by hand

• Quickstart
• Core concepts
• Run builder
• Read results

Sections

Section	Holds
Settings	Model providers, judges, agent keys, run hosts
Benchmarks	Every benchmark, and how to add one
Agents	Every agent, its proven ports, how to add one
Capabilities	Every capability, its port, how to add one
The standard	Ports, kits, setups, controls and proofs
CLI	Every command, by topic
Release notes	Highlights, changelog and installation notes by version
Architecture	Execution paths and files on disk
Metric field guide	Every metric and outcome; Reference → Metrics on the website

This folder

Path	Holds
settings/, cli/, benchmarks/, agents/, capabilities/	The sections above
releases/	Reviewed notes for each release
examples/new-benchmark/	A worked Harbor entry to copy
assets/	The screenshots these pages use
diagrams/	The figures these pages use, drawn by app/site/diagrams.py