KNOWWLearns from little data
Priors learned across millions of synthetic datasets carry it where your history is short: a new product, a new market, a new supplier.
Every business is trying to see tomorrow. Tree-based models make it slow.
LTFM is a foundation model for tables that change over time. Predict from your history, with uncertainty shown alongside every answer.
MOne new question, one new model↻Rebuilt when the world changes
Train once on millions of worlds. Infer on yours.
Prior-data fitted networks learn how structured data behaves from synthetic datasets. Shown new data, they find the pattern in a single forward pass.
01Train · once
synthetic datasets × millionstransformer
A transformer learns from millions of synthetic records: what trends, seasonality, noise and shocks look like.
02Infer · one forward pass
01 / 09 · Rising trendFinds the pattern, shows its uncertainty. No retraining.
Why LTFM
Four things today's predictive models don't do at once.
Learning-to-learn is already proven on tables. LTFM extends it to data that changes over time, in any domain.
Understands change
Reads each entity's history and the transitions between its states, not a snapshot.
Says how sure it is
Every score comes with a band. Less data means a wider band, and you can see it.
Adapts, no retraining
A new question is a new input. The model you validated is the model that answers.
Proof it works on tablesHollmann et al., Nature 637 (2025)
PFN, out of the box
2.8s
Experts tuning models
4h
A prior-data fitted network, out of the box, beat four hours of expert tuning. One forward pass is enough to replace tree-based models. How it works
What LTFM reads
Entities, their state, their history and every transition.
AI reads language, images and code. No LLM understands tables. LTFM is built for the rows and columns every business runs on: spreadsheets, databases, ledgers, logs.
Entity · 2026
J
F
M
A
M
J
J
A
S
O
N
D
Supplier S-114
Lane SH-3307└ run by S-114
Warehouse W-03└ fed by SH-3307
Product SKU-8812└ stocked at W-03
Customer CU-2291└ buys SKU-8812
Example · a supply-to-customer chainNormalWatchStressedFlagged▸ Transition└ Relation
Time-series
TimesFM, Chronos-2
Tabular
TabPFN → acquired by SAP
Relational
KumoRFM → NVIDIA
LTFM
entities + state + history + transitions
Existing models see a piece. LTFM sees the whole.
The questionWill this account default within 90 days?
| ID | Months on book | Utilisation · 12m | Payments · 12m | Income Δ | Probability · 90% band | Status |
|---|---|---|---|---|---|---|
| 38 | Utilisation 32%, 35%, 38%, 41%, 47%, 52%, 55%, 61%, 66%, 72%, 77%, 81% | -18% | 0.42 | High | ||
| 64 | Utilisation 21%, 20%, 22%, 19%, 20%, 21%, 20%, 18%, 20%, 19%, 21%, 20% | +4% | <0.01 | Low | ||
| 12 | Utilisation 48%, 62%, 41%, 66%, 52%, 58%, 44%, 69%, 51%, 60%, 57%, 55% | 0% | 0.02 | Low | ||
| 1 | Utilisation 40% | — | n/a | Abstain | ||
| 27 | Utilisation 70%, 68%, 66%, 62%, 60%, 57%, 53%, 50%, 48%, 45%, 42%, 40% | +9% | <0.01 | Low | ||
| 45 | Utilisation 60%, 62%, 65%, 68%, 70%, 74%, 77%, 80%, 83%, 86%, 88%, 90% | -6% | 0.48 | High | ||
| 19 | Utilisation 30%, 31%, 30%, 29%, 30%, 31%, 30%, 30%, 29%, 30%, 31%, 30% | +2% | 0.01 | Low | ||
| 8 | Utilisation 10%, 12%, 11%, 10%, 14%, 38%, 55%, 64% | -30% | 0.13 | Watch |
On timeLateMissedNo history
accounts scored7
Abstained1
Models trained for this table0
Forward passes1
Switch the table: same model, a different question, no retraining. Select a row to see why it scored the way it did. All rows and scores here are synthetic examples that show how LTFM reports results; they are not model output.
Product walkthroughs
Knoww platform demos
Two walkthroughs, from uploading a table to reviewing predictions and monitoring results.
Who it's for
Anyone with a table and a question about tomorrow.
Every industry runs on rows and columns that change over time. The same model answers each one's question. We start with financial services.
Financial services · firstWill this account default?Fraud · credit · underwriting · collectionsInsuranceIs this claim genuine?Claims · renewals · pricingSupply chainWill this shipment be late?Delays · supplier risk · lead timesRetail & inventoryHow much sells next week?Demand · churn · stock-outsManufacturing & maintenanceWhich machine will need maintenance next?Failure risk · downtime · maintenance costsPolicy & macroWill rates be cut?Rates · inflation · eventsYour tableWhat do you need to predict?Tell us the question →
Built to be trusted
Ready for high-stakes decisions.
Predictions that move money or stock need evidence. LTFM is validated like any model you run today, and it takes no decision until you sign off. We start with banks, the strictest review there is.
Validated like any model you run
Out-of-time backtests on your own history.
Every score has reasons
Plain-language reasons for every prediction.
Knows when not to answer
Abstains and falls back to your model.
Your data stays with you
Private deploy, full audit trail.
How every pilot runs
- Shadow mode
LTFM scores live cases. Your current process keeps deciding.
- Side by side
Both compared on the same cases.
- Validation sign-off
Your reviewers check backtests and reasons.
- LIVE
In production, with your current process as the fallback.
Roadmap · 24 months
Hire in 4 weeks. Prove in 16. Scale in 24 months.
A go/no-go gate at week 8: LTFM has to beat gradient-boosted models before we scale.
04812162024 MO
Hiring · 4 wks
resource the team
Feasibility · 16 wks
← go / no-go at week 8: beat GBMs
LTFM v1 · LSM + Intel.LTFM's internal state
→ grows into full LTFM
Pilotsshadow mode → live
LTFM v2 · full platformsame core, every industry
Open researchintake every 8 weeks
| = open-research intake every 8 weeks
Request a pilot
What do you want to predict?
Every pilot starts in shadow mode. LTFM predicts on your cases next to what you use today and changes no decisions. Tell us what you want to predict and we'll reply with a pilot plan.