Spectral Codex

Autonomous AI engineering that compounds. Spectral measures, optimizes, and improves complex AI systems against domain standards your team didn’t have to author.

Codex is how that system is built. It covers the design, the domain knowledge that grounds evaluation, and the patterns engineers follow in the codebase.

Spectral is for systems where agent count, interaction complexity, or evaluation surface have outgrown what manual tuning and hand-built evals can cover — especially multi-agent systems, where optimizing one agent shifts the ground under the others.

If you’re new

Read these in order. They take a cold reader from zero to oriented.

1. How Spectral Works — what Spectral does, and the two pillars (the platform, the world-model system) that make it possible.
2. Get Started — two end-to-end walkthroughs: an operator bootstrapping a world model; a customer earning their first agent performance card.
3. Domains — the domain knowledge that grounds evaluation. Tax-prep’s problem spaces, source materials, and rules live here.
4. System Design — foundations, agents, the world-model system, the platform pillar, operator surfaces, and topology.
5. Your First Feature — the contributor walkthrough. Build a feature end-to-end through every layer.

Looking for something specific

Browse the sidebar or search with Ctrl+K / Cmd+K.

• Glossary — canonical terminology; one-line definitions for cross-referenced concepts.
• Decisions — every ADR and its supersession graph. The “why” behind each choice.