A compiler for AI thought

Make the AI
show its work.

Eigenius is an open-source platform — a typed knowledge graph for AI-assisted scientific reasoning. Every claim records how it was justified, so a brilliant analysis and a confident hallucination can no longer look the same in prose.

See a worked example View on GitHub
The trust gap

AI is increasingly the author.
The reasoning chain is still invisible.

Drug-discovery pipelines, autonomy stacks, climate models, materials screens — AI agents are drafting more and more of the scientific and engineering record. The accountability surface science was built on is binary: either you trust the author, or you don’t use the result. At human authoring speed that trade-off was tractable. At LLM authoring speed it is not.

Consider the sentence “EIG_0291 was identified as a strong CDK2 inhibitor (IC₅₀ = 85 ± 5 nM, n = 6).” Whether drafted by a human author or by an AI agent, that sentence carries five distinct implicit assertions about measurement, statistics, vocabulary translation, decision rule, and sample attribution. None are visible in the published sentence. Each is load-bearing. Each is where a hallucination or unjustified assumption can hide while looking exactly like the correct article.

The structural answer

Every claim — human or AI — lands in one of four warrant categories.

The chain records how a claim is justified. Three warrant kinds are mechanically warranted — an instrument run, a deterministic computation, or a formal proof — and can be independently re-executed by anyone with read access. The fourth carries an authoritative warrant: a named declarer is on record, and the citation chain is examinable.

Mechanical warrants re-executable
Observed Instrumental

Raw measured data with unbroken instrument provenance.

Audit fingerprint Instrument run, time, operator.

Example: Six raw IC₅₀ readings from a Kinase-Glo plate run.

Derived Computational

Outputs of deterministic computation, rerunnable byte-identically.

Audit fingerprint Procedure, inputs, kernel that ran it.

Example: A t-statistic and p-value the statistics institution emitted.

Verified Formal-proof

Claim accepted by a machine-checked mathematical proof.

Audit fingerprint Proof artifact, checker, independent re-execution.

Example: A Lean-verified regulatory primary-endpoint result.

Authoritative warrant examinable
Declared Authoritative

Asserted by a named human, organisation, or policy document.

Audit fingerprint Declarer identity, citation, organisational authority.

Example: A 100 nM threshold cited from a methodology document.

The four categories are peers, not a hierarchy. None is more “true” than the others; they differ in what kind of warrant they carry. The chain doesn’t prevent an AI from declaring something false, but it does prevent the AI from passing a declaration off as a mechanical warrant.

Where to start

Three ways into the platform.

For scientists working with AI

Walk any claim — yours or an AI agent's — back to the raw data.

  • Every step from instrument to designation is a typed artifact
  • Counterfactual analyses become localised chain edits
  • FAIR and ALCOA+ alignment falls out structurally
Read the worked example → For AI developers

A type system your agent must justify against.

  • The kernel refuses to commit ill-typed reasoning
  • Hallucinations land as Declared evidence, not derivations
  • Audit closure becomes a tree-walk, not a re-derivation
Read the concepts → For platform engineers

Wire new AI reasoners into the chain through a modular institution architecture.

  • Add a new dispatch surface without touching the kernel
  • Content-addressed layers, deterministic by construction
  • Open source kernel + ESL surface + EigenQL queries
Read the docs →
Watch

Three minutes on what Eigenius is.

The infrastructure-for-truth pitch in video form. Additional explainers covering specific concepts (cross-vocabulary bridges, AI reasoning verification, kernel design) are embedded on the relevant Concepts pages, and the full playlist on YouTube ↗ walks through everything.

The platform pitch: how Eigenius makes scientific reasoning machine-checkable, so AI agents have to show their work instead of letting it hide in author prose.

Writing

Institutions — a three-part introduction.

A logic is four parts and one law. Signatures, sentences, models, satisfaction — and the condition that ties them together. A self-contained conceptual foundation, with no category theory required.

  1. Part IPart I — What Is a Logic?
  2. Part IIPart II — Translating Between Logics
  3. Part IIIPart III — Making Reasoning Checkable

Start with Part I All writing

Last updated · RSS

Research

The case in long form.

arXiv Preprint · cs.DB

Eigenius: A Typed Knowledge-Graph DBMS with Epistemic Stratification and Institution-Mediated Reasoning

A typed knowledge-graph DBMS that makes provenance a structural invariant rather than something reconstructed across subsystems: epistemic status is a commit-time check, institutions are strongly typed integration boundaries, and cross-system translations materialize into the graph as first-class resources. Recomputing a published Nature study end to end, all 52 derived conclusions hold from pinned data — surfacing four machine-checked discrepancies in the original.