Skip to content
Eigenius

A Compiler for AI Thought

Coming soon. This paper is in active preparation for submission to Communications of the ACM. We are not circulating the manuscript or abstract yet. This page will be updated with the submission draft, abstract, and reading guide as soon as the paper is ready to share.

What the paper will cover

A systems contribution framing the Eigenius platform as a compiler for AI thought: a typed, content-addressed knowledge graph in which every step of an AI agent’s reasoning becomes a first-class, machine-checkable artifact. The kernel acts as the type-checker the agent must produce against; the type discipline forces hallucinations to fail to commit or land visibly as authored declarations.

The paper will introduce the four-warrant taxonomy (Observed, Derived, Verified, Declared), the role of institutions as modular dispatch surfaces, the domain-bridge primitive as the declarative first stage of crossing between vocabularies, and the drug-screening example as the end-to-end demonstration.

In the meantime

The conceptual material the paper will cover is available now in narrative form on this site:

  • The Concepts section walks through the four-warrant taxonomy, domain bridges, institutions, and the justification-logic grounding.
  • The Drug-screening example is the same worked example the paper will use.
  • The Docs section covers the surface languages, the query language, and the platform — the layer the paper sketches as “the compiler.”

Open source

All code that will be referenced in the paper is at github.com/eigenius/eigenius.