Research infra every researcher can run

Most AI research capabilities are built for large-scale compute clusters and unlimited token quotas. Autonomio designs its open-source research tools for use from an ordinary laptop, so that a researcher can conduct an investigation powered by frontier intelligence even if they don’t have access to extraordinary resources.

The laboratory

Autonomio is an independent, non-profit AI research laboratory in Finland, founded in 2014. Its purpose is to make scientific research capabilities accessible to more researchers.

History

  1. Early 1990s

    Autonomio’s founders became interested in artificial intelligence.

  2. 2005–2007

    Work on relations between words and phrases in ordinary text led to an international patent filing.

  3. 2014

    A prototype built from Wikipedia raised the question of whether a machine could find more structure in the same text by trying different mathematical representations.

  4. 2018

    Talos gave researchers a way to automate deep learning experiments using their own models and data, and introduced energy consumption as a measurable training outcome.

  5. 2019

    The Automated Research programme described how a future system could investigate a question stated in ordinary language.

  6. 2026

    Autonomio introduces Inferons, a proposal for highly specialised models with a common protocol for communication.

LOAD"MATROSKA",8,1

Frontier AI on the computers researchers already use

Project Matroska is Autonomio’s direct challenge to a paradigm in which only a few frontier AI labs can afford to train models with the intelligence and capabilities of human researchers.

  1. Stage 1 of 5

    A measurement rig

    The programme began in 2026 with a measurement rig on an ordinary laptop.

  2. Stage 2 of 5

    Data preparation

    The rig’s measurements guide reductions in the computation that data preparation requires, and researchers select which data to use for training.

  3. Stage 3 of 5

    Model training

    The same measurements guide reductions in the computation of model training, and researchers choose how the models are trained.

  4. Stage 4 of 5

    Inference

    Inference is reduced in the same way, and researchers choose how the inference engines are used.

  5. Stage 5 of 5

    Binary logic circuitry

    The programme concludes by 2030 with the release of self-assembling binary logic circuitry that runs on conventional hardware.

A pixel matryoshka whose dolls open one stage at a time
Stages of the programme

From question to thesis

Autonomio tools bring question formulation, instrument development, systematic literature review and thesis formulation into one end-to-end pipeline that any researcher with an ordinary computer can run.

Poise

Builds and tests any software instrument the investigation requires.

Canary

Produces the systematic review, with the source passages for its findings and a record of unresolved questions.

Galley

Develops the thesis into a pre-publication draft and verifies its citations.

  1. Instruments and their tests
  2. The review, its source passages and its open questions
  3. The draft and its verified citations

Resolvent

Structures all the outputs into one shared world-model.

We started by playing with computers more than four decades ago, and never stopped.