Inserloft Research001Active

Project Proteus

AI for the next generation of biological discovery.

Proteus explores how machine intelligence can help researchers understand, model and design biological systems, from proteins and molecules to broader discovery.

Engraving of a protein ribbon.

Why Proteus

Discovery is a search problem.

Biology is a vast space of possibilities. The challenge is deciding what to explore next.

A research system should not try to know everything. It should help determine what is worth knowing next.

Engraving of a seahorse, the hippocampus of Greek myth.
  1. 01Observe

    Learn from sequences, structures, interactions and measurements.

  2. 02Prioritize

    Rank possibilities so attention goes where value is highest.

  3. 03Explore

    Generate and investigate candidates to narrow the search.

  4. 04Learn

    Feed results back to improve the next search.

The goal is not to predict everything. It is to make the next experiment more informative.

Search spaceBiological possibility
ObjectiveBetter scientific decisions
LoopObserve, prioritize, explore, learn

Philosophy

Intelligence should expand scientific possibility.

AI should extend scientific reasoning, not replace it. The standard is whether it produces better science.

Evidence over appearance

Judge the work by reproducible results and validation, not by demos.

Exploration before certainty

Make room for unknowns in early research.

Models serve the research

Change architectures when experiments show a better path.

Research lines

Intelligence meets biology.

A research program with several connected directions.

  • A

    Protein intelligence

    Learning from sequences, structures and functions.

  • B

    Generative biology

    Generative models that propose and navigate design spaces.

  • C

    Discovery systems

    Connecting models, data and evaluation into repeated research cycles.

  • D

    Biological representation

    Making diverse biological data computationally usable together.

  • E

    Candidate search

    Navigating large candidate sets under scientific constraints.

  • F

    Research infrastructure

    Data pipelines, evaluation and compute.

Model foundation

Caelis­Neural­Base 3 is now part of the Proteus foundation.

Inserloft Research released CaelisNeuralBase 3 (CNB 3), a public protein foundation model for design and discovery.

CNB 3Model: CaelisNeuralBase 3
985.4MParameters, CNB3 architecture
MITLicense, public on Hugging Face
FoundationRole: protein-focused systems

Scope

For a better world. For a better life.

Proteus is not tied to one disease or one target. It builds technology that makes biological discovery more tractable.

Human health

Medicine discovery, disease biology and therapeutic candidates.

Animal health

Veterinary medicine and animal disease research.

Nature

Understanding ecological systems and protecting living systems, where supported by science.

Long-term view

An open-ended scope, set by evidence and feasibility.

How it works

Think. Generate. Evaluate. Learn.

A continuous loop: every output is evaluated and fed into the next iteration.

  1. 01Observe

    Collect biological data, results and context.

  2. 02Represent

    Turn information into model-ready representations.

  3. 03Generate

    Produce predictions, hypotheses or designs.

  4. 04Evaluate

    Test against criteria, simulations or experiments.

  5. 05Learn

    Feed results back into the system.

  6. 06Iterate

    Repeat until the question sharpens.

System

A research stack, not a single model.

An evolving system of models, data, evaluation and tooling.

Proteus research nodeActive
$ initializing proteus.research
$ loading biological representation layer
$ connecting model + data + evaluation
  search space: biological systems
  objective: accelerate discovery
  status: ACTIVE
_ awaiting next experiment

Scientific discipline

Fast research still needs slow truth.

Evaluation is part of the architecture. Predictions, simulations and experimental results stay distinct.

S/01

Measure

Meaningful evaluations, not impressions.

S/02

Compare

Benchmarks against credible baselines.

S/03

Validate

Computational results connected to stronger evidence.

Responsibility

Capability without care is not progress.

Biological design systems require scientific responsibility and awareness of misuse.

R/01

Scientific honesty

Preliminary results are never presented as established.

R/02

Responsible access

Outputs are shared to maximize legitimate value and limit misuse.

R/03

Human judgment

AI remains a research instrument; experts make the key decisions.

Building in public

Research should leave a trail.

The work becomes more transparent as it matures and becomes reproducible.

What may become public

Research notes, articles, model information, benchmarks, datasets and tooling, where appropriate.

What comes first

Scientific quality, reproducibility and responsible disclosure.

Why

Others can build on research that documents what worked and what failed.

Philosophy in practice

Build. Test. Measure. Repeat.

Intentionally unfinished, and expected to change.

Current phaseResearch and system development
FocusAI-assisted biological discovery, protein intelligence, and generative design
ScopeHuman health, animal health, and broader biological discovery
MethodBuild → test → measure → learn → repeat
OrganizationInserloft Research

The future

What comes next may not look like today’s models.

The long-term goal: AI systems that are useful partners in difficult biological research, grounded in evidence.

Model 001

Caelis Neural Base 1.

CNB-1 is the first experimental Proteus model. Early tests produced structured biological outputs, including a PET-hydrolase candidate of about 350 amino acids.

Engraving of a radiant sun over a calm horizon.
ModelCaelis Neural Base 1 (CNB-1)
StatusExperimental research model
First targetPET-hydrolytic enzyme candidate
EvaluationProteinGym, TAPE Fluor, TAPE SecS, CATH Accuracy, FoldSeek
ExperimentStructured biological generation
NextStructure, function, validation
M/01

Structured outputs

Interpretable sequence candidates.

M/02

Biologically informed seed

A seed tied to a signal-peptide motif in natural PETases from Ideonella sakaiensis.

M/03

Computational benchmark

Benchmark results measure tasks, not biochemical activity.

Project Proteus, Inserloft Research.

Back to Research