The prediction layer for
people and AI agents.
Drop in a dataset and JITM will train, evaluate, explain and host a model for you. No ML expertise needed.
Free to start. No card required. Your data never improves anyone else’s model.
Highest: espresso beans, 92% chance of stockout by Friday. Model is 87% accurate on real outcomes.
{
"at_risk": 14,
"top": {"sku": "BR-1042", "chance": 0.92},
"top_factors": ["recent_sales", "lead_time", "stock_level"]
}
Member of NVIDIA InceptionAccuracy is a record,
not a claim.
Most tools show you a score from the day the model was built. JITM keeps checking its predictions against what actually happened, and shows you the record, gap and all.
How to read it: in this example, predictions marked 70% came true 68% of the time. The two-point gap shows how confidence compares with reality. Dots on the dashed line indicate a match. These are illustrative values, not a live model’s results.
Real questions, answered from your data.
Illustrative examples. Your numbers come from your data.
Made for people and AI agents.
Upload a dataset
Drop a CSV, JSON or Parquet file at the top of this page, pick the column you want to predict, and JITM does the rest: column analysis, feature engineering, training, evaluation, and a live endpoint you can call.
Free account, no card required.
Start here ↑Give your agent a brain
Connect Claude, ChatGPT, or Cursor to mcp.jitm.ai and your LLM becomes a working data scientist: it studies your data, trains the models, reads the metrics, and answers with live predictions.
No tokens or config files needed. OAuth signs you in automatically.
Start free. Grow into the rest.
Build it yourself
Upload your data or point your agent at us. Train strong models, serve live predictions, and see exactly why each one was made. This is the part you are looking at, and it is free to start.
Start free →And it stays right
Real outcomes flow back to the model that predicted them. When reality shifts, Evergreen tests a challenger and promotes it only once it proves better, and only once you approve.
How it works →Data in. Predictions out.
Three steps between your raw data and a production-ready prediction API.
Upload your data
Drag a file onto the page. We detect schema, missingness, quality issues, and likely targets, so you can make an informed choice before training.
Models train & compete
Automated feature engineering, task selection, fast training, and validation. No knobs to turn: we handle the hard parts.
Baseline in secondsGet your endpoint
A live API, an auth token, and feature importance rankings. Integrate your model into an agent, app, or workflow.
Know exactly what drives your predictions.
Every model includes feature importance rankings, so you can see which inputs matter most overall. Per-prediction SHAP explanations are in development.
Most models rot.
Yours stay evergreen.
Every real outcome flows back to improve the model that predicted it. When reality shifts, Evergreen tests a challenger and promotes it only when it proves better, and you approve. Every generation is logged and auditable.
Explore JITM Evergreen →Saturday was forecast at 520 units. Reality said 495. That difference is not an error to hide: it is tomorrow’s training data.
Every industry. Every agent.
When any AI agent can build its own predictive models, everything changes.
If it has data and needs predictions, JITM.ai powers it.
Simple, transparent pricing
Start free. Pro and Scale are invite only while we onboard teams by hand.
- MCP access (13 tools)
- Automated feature engineering
- Time-aware validation
- Leakage detection
- Sub-25ms predictions
Free
Try your first prediction model.
- Train and host one model
- See which features matter
- Connect through the API or MCP
All features View details
- 1 model
- 50 MB uploads
- 50K rows
- 20 API calls/hr
- XGBoost gradient boosting
- Accuracy, F1, ROC-AUC, RMSE
- Feature importance rankings
- No deep training
- No ensembles
Pro
Build models for everyday production.
- Deeper training with multiple model families
- Track outcomes and model drift
- Guided onboarding
All features 1 in development
- 25 models
- 1 GB uploads
- 5M rows
- 1,000 API calls/hr
- Batch predictions (100 rows)
- Everything in Free
- Deep training on your full dataset
- 4-family ensemble (XGBoost + LightGBM + GLM + TabM, plus CatBoost when useful)
- 5-fold cross-validation with 2-seed averaging
- Stacked meta-learner + hill-climb ensemble selection
- Outcomes endpoint: send back what really happened
- Drift monitoring + rolling skill score vs baseline
- Guided onboarding
- In build
- SHAP explanations per predictionSoon
Scale
Keep models learning as your data changes.
- Everything in Pro, plus Evergreen
- Approve challenger models before promotion
- Generation history and priority support
All features 3 in development
- 200 models
- 10 GB uploads
- 50M rows
- 10,000 API calls/hr
- Batch predictions (10K rows)
- Everything in Pro
- jitm evergreen
- Champion/challenger retraining with your approval
- Generation audit log
- 60-trial hyperparameter search
- Priority support
- In build
- 10 Evergreen-active models includedSoon
- Model lifecycle eventsSoon
- 3 seatsSoon
Additional Evergreen-active models are billed per model. API calls above the hourly allowance are metered.
Enterprise
A tailored setup for your organisation.
- Everything in Scale
- Hands-on onboarding and a named contact
- Custom terms and service agreement
All features 3 in development
- Custom model, upload and row limits
- Custom API volume terms
- Everything in Scale
- jitm evergreen
- Hands-on onboarding
- Named contact and SLA
- Custom terms, data rights and DPA
- In build
- Custom drift policies and retrain limitsSoon
- Unlimited seats, SSO, roles and permissionsSoon
- Audit logsSoon
Priced per engagement and invoiced directly by Hannai Labs Ltd.
Where it lives, who touches it, when it expires. No badges we have not earned.
Questions & answers.
Everything you need to know about turning your data into live predictions.
Latest from the team
Product updates, engineering deep dives, and announcements.
Your agent now builds a model team, not just a model
JITM.ai Pro now trains four model families in parallel and learns how to combine them into a single prediction endpoint. Connect any MCP-compatible LLM, hand it a CSV, and let it build the model team for you.
Read more →One URL, zero tokens: OAuth comes to JITM.ai
Connecting an AI agent to JITM.ai now takes exactly one step: paste the URL. OAuth handles the rest. No tokens to generate, no config files to edit, no secrets to rotate.
Read more →Three expert algorithms, time-aware splits: Phase 2 gets smarter
Every Phase 2 model on higher tiers now draws from three different learning algorithms and combines the best from each. And when your data has a time dimension, JITM.ai now respects it automatically, so models don't learn by peeking at the future.
Read more →New to ML? We'll help
you get up and running.
You don't need to be a data scientist to get a model into production. Tell us what you are trying to predict and we will point you to the right approach, help you get your first model trained and live, and answer whatever comes up along the way. Every paid plan comes with onboarding from a human, not a help article.
- Advice on what to predict and how to frame your data
- Help getting your first model trained and live
- Guided onboarding on Pro, hands-on onboarding on Enterprise