Skip to content

Platforms

Three live platforms, one discipline.

Everything on this page is live today, not a roadmap. Each platform is built the same way: every number traceable to its source, every engine named, and data you can export or self-host.

MATFLOW · CAPRI · CELARON CHAT

The platforms

Pick the one that matches your work.

Three domains, one engineering standard: transparent evidence, real engines, and no lock-in.

Matflow

Scientific R&D

Physics-validated AI for batteries, formulations, and materials.

Experimental rows in, ranked candidates out. Matflow turns sparse lab data into working models, cost-aware candidate rankings, and publication-grade dossiers — with every number traceable to its source run, engine, and evidence class.

  • 0.41–0.95

    Held-out R² across public benchmark datasets

  • 0.95

    Leave-family R² on the bundled demo corpus

Explore Matflow

CAPRI

Precision Oncology

Evidence-ranked variant interpretation, in one workbench.

Cancer Actionability & Precision Response Interpreter. CAPRI unifies twenty-nine biomedical sources into a single workbench — normalized variants, AMP/ASCO/CAP-style Tier I–IV actionability, structure-aware drug context, cohort analytics, and de-identified tumor-board summaries.

  • 510,504

    Variants in the live catalogue

  • 1,084

    Genes in the live catalogue

Explore CAPRI

Celaron Chat

Private AI chat

A private AI workspace you actually control.

Celaron Chat is a self-hosted workspace for everyday AI work — conversations, documents, and images on infrastructure you control, with the model options you choose.

  • Open WebUI

    Open-source foundation, v0.8.12

  • Self-hosted

    Runs on infrastructure you control

Explore Celaron Chat

What we are exploring

Nothing to announce yet.

We are exploring agentic coding tools that run with your own models on your own infrastructure. It is early work: there is no waitlist, no launch date, and no product page — and we will not present it as one until it is real.

Tell us what you would wantFollow along on GitHub

Evidence discipline

Every number shows its work.

Each value our platforms produce carries an evidence class and a link back to the source run or engine behind it.
How we handle trust
  • MEASURED

    Raw experimental or instrument result.

  • COMPUTED

    Output of a real engine run — PyBaMM, M3GNet, pycalphad.

  • PREDICTED

    Registered model output with calibrated uncertainty.

  • EXTRACTED

    Pulled from literature, flagged for review.

  • HYPOTHESIS

    A generative proposal, never presented as fact.

  • DEMO

    Synthetic or illustrative — never billed, never disguised.

Not sure which platform fits?

Tell us what you are working on. We will point you to the right tool — or say honestly that none of them fit.

bassil@celaron.comBrowse details of Matflow