Get started · ~15 minutes, no account

Run the notebook.
Commit a chain.
Audit a claim.

Eigenius runs on your laptop. The walkthrough is a notebook; each cell commits a typed artifact to the chain, and the kernel type-checks the composition as it goes. By the end you have a machine-checkable audit trail from six raw plate-reader readings to a published-style designation.

Install with Docker View on GitHub
See it

The notebook is the platform.

Every cell commits a typed chain artifact. The right pane surfaces the audit trail the kernel just type-checked — not a rendered summary, the actual chain. Below is the drug-screening walkthrough partway through.

Eigenius notebook UX: a left pane of cells progressively committing typed chain artifacts (Observed, Derived, Declared), with a right pane showing the live chain state — types resolved, AutoOnLoad cascades fired, the type-checker's verdicts visible.
Install

Three steps to a running chain.

The end-to-end demo runs entirely in Docker — no Rust or Deno on the host required. If you'd rather build from source, follow Platform §2 — Installation instead.

  1. 01

    Clone and start the stack

    git clone https://github.com/eigenius/eigenius
cd eigenius
docker compose up -d

    Requires Docker Desktop or Docker Engine + Compose v2. The first run pulls images and seeds the layer chain — allow a couple of minutes.

  2. 02

    Open the notebook

    http://localhost:8080/notebooks/

    No login. Everything runs on your laptop. The orchestrator serves the React notebook and proxies EigenQL/ESL into the kernel over gRPC.

  3. 03

    Run the drug-screening walkthrough

    Open examples/stats-and-reasoning.json and run cells top to bottom. About five minutes end-to-end. Watch the AutoOnLoad cascades fire as the statistics institution emits its verdict and the reasoning institution composes the certificate.

    Read the walkthrough first Platform §14 — Notebook reference

You just did

The chain, now machine-inspectable.

By the time the last cell runs, you have committed:

  • Observed six raw plate-reader IC₅₀ measurements, each carrying its instrument provenance.
  • Derived a deterministic statistical verdict, recomputable by anyone with read access to the chain.
  • Declared a chain-resident methodology — the polymorphic domain bridge from statistical claim to compound classification, on the record as its own citable artifact.
  • Derived the final conclusion as a typed proof term the kernel checked at commit time. No prose, no hidden assumptions.

A reviewer (or another AI agent, or a regulator) can now walk back from the conclusion to the raw data in finite steps, re-execute any deterministic step, and refute any premise without disputing the rest.

Where to next

Three directions from here.

Learn the concepts

The framework's vocabulary in narrative form.

  • The four-warrant taxonomy: Observed, Derived, Verified, Declared
  • Domain bridges: cross-vocabulary translation as data
  • Institutions: how runtimes cooperate without sharing a language
Read the concepts → See more examples

Worked end-to-end walkthroughs.

  • Drug screening: the statistics + reasoning pair
  • Kinase potency mapping: five Julia institutions in concert
  • Lean-verified primary endpoint (planned)
Browse examples → Read the reference

Surface languages, queries, and the platform.

  • ESL — declare ontologies and typed programs
  • EigenQL — query the layered Eigon graph
  • Composition, formula language, platform internals
Open the docs →
Help

Stuck? Found a bug?

Project tracking lives on GitHub. Use Discussions for open-ended questions and design conversations; file an Issue with the smallest reproduction you can manage when something is broken. For private deployments and commercial questions, see the contact note in the repo README.

Join the discussion File an issue View on GitHub