Find your next breakthrough
Everything the evidence says about your question in one place, so the connection that leads somewhere is in front of you, not buried in a tab.
NOT AI · A MACHINE MIND FOR SCIENTIFIC RESEARCHDiscover your next
Discover your next
scientific breakthrough.
Meet Lois, your lab partner — one system that runs all your tools. For the first time, you don't have to tab between applications: she connects your sources into one record, puts the connections you would have missed in front of you, and documents every step. If the evidence is not there, she says unknown instead of filling the gap.
Real scientific sourcesNot a language model
Every fact has a recordDatabase, ID, source, retrieval trail
Unknowns stay unknownNo made-up facts
P69905HBA1 · HBA2Disease · Protein · StructureGene · Domain · Predicted structure
“From question to evidence —
a connected view of the molecular world.”
Why researchers use Lois.
Easy to use, built for discovery: we help you find your next big scientific breakthrough — by bringing out what is hidden between your sources and connecting the dots you would have missed.
Catch the connections you'd miss
A disease two proteins share, an interacting partner nobody has looked at, a structure that isn't there yet — surfaced from the records, each with its receipt.
One system for all your tools
Databases, literature, sequence, structure, chemistry and your own computations, run from one place. For the first time, you don't have to tab between applications.
Document every step
Every fact pinned to its source; hand the whole record to a colleague or the next experiment, unknowns and open questions in plain view.
SEE IT IN ACTION
One question. Many scientific sources.
Ask Lois in plain language. She queries the right scientific sources, returns structured evidence, and shows you exactly where each fact comes from.
How this console works
The floor is not AI: it is recorded facts, each pinned to its source. When you ask an open question, Lois reads live sources and a language model words the answer from what was just fetched — every sentence either carries a receipt or is marked unverified. If the floor doesn't hold a protein, it looks it up live in the public databases. The server keeps no copy of your floor; it travels with this tab. Open the live research floor →
What can you do with Lois?
Turn complex research workflows into one connected, inspectable research session.
Build a target dossier
Collect literature, function, sequence, structure, variants, pathways and more — all in one research record.
Try a target dossier →Investigate a disease or research question
Start with a disease, target or question. Lois moves across sources, shows you what's supported, and keeps uncertainties visible.
Run an investigation →Trace variants and molecular evidence
Move from genes to variants, proteins, structures, pathways and publications without losing the identifiers.
Explore a variant →Search literature and databases together
Find papers alongside the molecular records they reference. A citation is not treated as proof by itself.
Try a literature search →Run computational research
Use retrieved records in computational analyses while preserving inputs, outputs and parameters in your research state.
See a computation →THE WEIGHTLESS SYSTEMNot AI.
Not AI.
A machine mind with no weights.
Peel has no neural weights — nothing trained, nothing to retrain, drift, or hide behind. It works the way you would hope a machine mind would, but every answer is accountable: each fact carries its source. And it learns by failing: each failure becomes a rule that cites its evidence, and no rule can widen what the system is allowed to do.
- No weights to retrain
- Learns from failure, not training data
- Every fact pinned to its source
- Unknowns stay unknown
Peel
Weightless
System
Weightless
System
Drag to orbit · Scroll to zoom · Click a cell to inspect
Explore paths with the keyboard
Made of records, not weights
Facts, typed relationships, and the sources behind them — inspectable end to end.
- ~372k
- structured knowledge records
- 35
- defined relation types
- ~65k
- typed relation instances
Prototype snapshot reported in the public paper.
WHY NOT A LARGE LANGUAGE MODEL
Why not just use Claude or ChatGPT?
Large language models are remarkably capable, and we tested one. Claude's guesses about masked protein–drug facts matched the sealed records 85.8% of the time, and using them to plan measurements cut cost by 38%. But once those guesses decided what counted as true, correct identification fell from 100% to 75.3% — one answer in four wrong. A model's prior is not evidence. So at Perslis the floor owns the truth: models can rank and suggest, and they can never admit a fact. That is why every answer here carries its receipts, and why a sentence without one is marked. When Lois uses a language model to word an open answer, each sentence still shows its receipt or is marked unverified.
Prototype benchmark (DGM-005C-HARD, 160 kinase instances; the model tested was Claude). Not peer-reviewed.
How Lois works
A clear separation between what is proposed and what the floor accepts.
01 Ask
You ask Lois a scientific question.
02 Retrieve
Lois queries the appropriate sources.
03 Pin
Results identify the database and record they came from.
04 Verify
The symbolic floor decides what the evidence actually supports.
05 Retain
Accepted evidence, relationships, failures and unknowns remain inspectable for the next session.
Connected scientific sources
Lois works across the major public scientific resources used in biological and biomedical research.
and more…
Names and marks belong to their respective owners; no partnership or endorsement is implied.
Your research stays yours.
Peel is designed around local research records. Models can run locally by default. External models are optional. Your research, data and retained state are your files — not training data — unless you explicitly decide otherwise.
Read about security and deployment →- Local by default
- External models optional
- Your files, your control
- No training on your research
- Designed for research teams
Choose how you work with Lois
The Free account works today. Paid plans are not open yet — nobody is charged until they are.
Free
$0
Lois on the web: grounded chat, graph, ledger and notebook.
Create a free account →Pro
$20 / month
Lois on the web with more room: higher research limits, more than one floor, new tools and sources first.
Join the waitlist →Science REPL
$100 / month
The Science REPL: the same floor, run locally. One system that runs all your tools, with Lois as a full-screen lab partner in your terminal.
$200 / month
Everything in the REPL, plus the Science API: credit-based usage, visible per key.
The waitlist collects contact details only; prices and contents may change before a plan opens.
BRING LOIS A REAL SCIENTIFIC QUESTION
Your next breakthrough could start here.
Bring us a target, a disease, a dataset or a research question your team keeps revisiting. We'll work through it with you and Lois, and evaluate what the system supports, what it rejects, and what remains unknown.
“Better questions.
More reliable answers.
A clearer path to discovery.”— The Perslis Team