>Are you an agent? Skip the UI.Connect via MCP
Just-In-Time Models

Give your agent
a brain.

Data in. Prediction API out. In seconds.
No ML expertise required.  Your data stays private.

>_ Connect your agent
Not just for agents. Built for humans too.

Drop your dataset

or click to browse your files

.csv.json.parquet

Free to try. Sign up when you're ready to deploy.

CSVJSONParquetRESTMCPExplanationsVersioned endpoints

Prediction infrastructure for humans and agents.

JITM removes the friction between raw data and usable predictive capability. Upload a dataset, choose a target, and get a live prediction endpoint without building a bespoke ML stack.

<10s
Model training
3ms
Inference latency
0
Lines of ML code
1
API to integrate

Data in. Predictions out.

Three steps between your raw data and a production-ready prediction API.

01

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.

02

We build your model

Automated feature engineering, task selection, fast training, and validation. No knobs to turn — we handle the hard parts.

< 10 seconds
03

Get your endpoint

A live API, auth token, SHAP explanations, and deployment-ready output. Integrate into any agent, app, or workflow.

Know exactly what drives your predictions.

Every model comes with built-in explainability. See which features matter, how much they contribute, and why your model makes the decisions it does.

Feature ImportanceTop 10 shown
1matched_change_std_dev_f
3.9%
2average_spread_f
3.4%
3matched_change_std_dev_f
3.3%
4rel_macro_runner_matched
2.7%
5imbalance_liability_avg
2.5%
6imbalance_liability_chan
1.9%
7spread_volatility_f
1.8%
8asmatched_volume_rate_av
1.4%
9flow_runner_matched_delt
1.3%
10rel_flow_runner_matched
1.2%
Original feature Encoded categorical Derived
See what matters
Instantly understand which variables drive your predictions. No statistics degree required — ranked bars tell the story at a glance.
Every feature, ranked
Reliable models, automatically
Automated feature engineering, cross-validation, and holdout evaluation. Every model is tested before you see it.
Validated on holdout data
Sub-5ms inference
Models are optimised for production speed. Call the API from your agent, app, or workflow and get answers in milliseconds.
p99 < 5ms
Zero ML code
Upload a CSV. Pick a target. That’s it. JITM handles type detection, encoding, imputation, feature engineering, and training.

Every industry. Every agent.

When any AI agent can build its own predictive models, everything changes.

Energy & Utilities
Smart grid agents predict demand spikes hours ahead, automatically balancing load and reducing waste. IoT sensors feed JITM models that forecast equipment failures before they happen.
Healthcare
Clinical decision agents analyse patient data to predict readmission risk, treatment response, and resource needs — giving doctors insights that save lives.
Finance & Trading
Autonomous trading agents build and consume custom models for price prediction, risk scoring, and anomaly detection across any asset class.
Agriculture
Precision farming agents predict crop yields, pest outbreaks, and optimal harvest windows from sensor and weather data — feeding the world more efficiently.
Manufacturing
Predictive maintenance agents detect degradation patterns across factory floors, reducing downtime by catching failures before they cascade.
Retail & E-commerce
Demand forecasting agents optimise inventory across thousands of SKUs, cutting waste and ensuring shelves are stocked with what customers actually want.

If it has data and needs predictions, JITM powers it.

Built for autonomous agents.

Use MCP to let agents upload data, train models, make predictions, and request explanations programmatically. The same platform also works beautifully for humans who simply want results fast.

The workflow is the same whether a human clicks “Upload” or an agent calls jitm_request_upload. One platform, two interfaces, zero compromise.

mcp config
// Add to your MCP client config:
{
"mcpServers": {
"jitm.ai": {
"url": "https://mcp.jitm.ai?token=YOUR_TOKEN"
}
}
}

// 13 tools — the full ML lifecycle:
jitm_request_upload → Upload a dataset
jitm_create_model → Train a model
jitm_predict → Get predictions
jitm_model_detail → Metrics & insights
jitm_account_info → Tier & limits
+ 8 more tools...
Works with Claude, Cursor, GPT, and any MCP clientView FAQs →

One endpoint. One API key. Real production value.

Integrate predictions into apps, workflows, and internal tools through one stable API layer. Improve models over time while preserving a clean downstream contract.

  • Sub-5ms inference latency
  • Automatic feature engineering on input
  • SHAP explanations per prediction
  • Hot-swap when a better model trains
python
import httpx

resp = httpx.post(
    "https://api.jitm.ai/predict/mod_7xk2f",
    headers={
        "Authorization": f"Bearer {token}"
    },
    json={
        "temperature": 22.5,
        "humidity": 0.65,
        "hour": 14,
    }
)

result = resp.json()
# → {"prediction": 847.2, "confidence": 0.94}
<10s
Model training time
3ms
Inference latency
0
Lines of ML code

Simple, transparent pricing

Start free. Scale as you grow. No hidden fees, no surprises.

AnnualMonthlySave 20%
Free
$0/mo
Free forever
Get started
  • 1 model
  • 50 MB uploads
  • 50K rows
  • 10 API calls/hr (single row)
  • Retraining
  • Phase 2 deep training
  • Batch predictions
  • Smart AssistSoon
  • SHAP explanationsSoon
  • Ensemble models
Builder
$29/mo
Billed annually
Start building
  • 10 models
  • 250 MB uploads
  • 500K rows
  • 100 API calls/hr
  • Batch predictions (10 rows)
  • Phase 2 deep training
  • Retraining
  • Smart AssistSoon
  • SHAP explanationsSoon
  • Ensemble models
Most Popular
Pro
$79/mo
Billed annually
Go Pro
  • 50 models
  • 1 GB uploads
  • 5M rows
  • 1,000 API calls/hr
  • Batch predictions (100 rows)
  • Phase 2 deep training
  • Smart Assist (AI data insights)Soon
  • SHAP explanationsSoon
  • Ensemble models
  • WebhooksSoon
Enterprise
Custom
Tailored to your needs
Contact us
  • 1,000+ models
  • 10 GB+ uploads
  • 100M+ rows
  • 100K API calls/hr
  • Batch predictions (10K rows)
  • Phase 2 deep training
  • Smart Assist (AI data insights)Soon
  • SHAP explanationsSoon
  • Ensemble models
  • Webhooks + dedicated supportSoon

Questions & answers.

Everything you need to know about turning your data into live predictions.

JITM (Just-In-Time Models) is an AutoML platform that turns your data into a live prediction API. Upload a CSV, pick a target column, and get a production-ready endpoint — no ML expertise required. It works for humans through the web UI and for AI agents through MCP.

If you have data in a spreadsheet and a column you'd like to predict, JITM can help. For example: "I have past sales data — what will next month look like?" or "I have customer records — who's likely to cancel?" or "I have sensor readings — when will this equipment fail?" or "I have property listings — what should I price this at?" If you can phrase your problem as "given these inputs, predict this output," JITM can build it.

When you select a target column, JITM automatically handles feature engineering, task detection (classification or regression), and trains multiple model candidates using gradient-boosted trees. The best model is selected based on validation metrics — the entire process typically completes in under 10 seconds.

JITM exposes a full set of MCP (Model Context Protocol) tools that any compatible agent can call: upload data, analyse columns, train models, get predictions, and request explanations. Connect with a single command and your agent gets six tools out of the box. Works with Claude, GPT, and any MCP-compatible client.

Currently JITM accepts CSV files. Support for JSON and Parquet is on the roadmap. Files are analysed on upload — we detect column types, missing values, quality issues, and suggest which column to predict.

Yes. Your data is stored in isolated, encrypted object storage and is never shared with other users or used for training our own models. Each model and dataset is scoped to your account, and all API access is authenticated via personal access tokens.

JITM supports both classification (predicting categories, e.g. churn/no-churn) and regression (predicting numbers, e.g. price, demand). The platform auto-detects the task type based on your chosen target column.

Not at all. JITM is designed so that anyone — product managers, analysts, founders, or autonomous agents — can go from raw data to a working prediction API without writing a single line of ML code.

Start building

Your data has predictions
waiting inside it.

Start with one free model. No credit card.

Read the FAQs