A stateless forecasting API your agent calls mid-workflow.

Ten objects, two APIs, one mental model.

The forecasting API turns a dataset into a queryable model. The outcome API turns your agent's own execution traces into a model of what succeeds and what fails. Both share the same auth and base URL.

The full surface: two families of routes.

Everything is versioned under /v1.0 and authenticated with a bearer license_token from register. The forecasting routes build and query models; the outcome routes model how your agent performs.

Real calls, real payloads.

A demand predict with its response, a what-if simulate, and the Python quick-start. Same SKU-4521 example throughout, copy, swap the token, run.

curl -X POST https://cortex.phaeth.com/v1.0/models/mdl-demand-v1/predict \
  -H "Authorization: Bearer $LICENSE_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "inputs": [{
      "sku_id": "SKU-4521",
      "price": 29.99,
      "promotion_active": false,
      "day_of_week": 2,
      "competitor_price": 32.50,
      "weather_temp_c": 24.0
    }],
    "horizon": 7,
    "include_confidence_interval": true
  }'

{
  "model_id": "mdl-demand-v1",
  "predictions": [{
    "entity": "SKU-4521",
    "forecast": [
      {"date": "2026-06-23", "value": 342, "lower": 295, "upper": 389},
      {"date": "2026-06-24", "value": 318, "lower": 271, "upper": 365},
      {"date": "2026-06-25", "value": 356, "lower": 309, "upper": 403}
    ]
  }],
  "confidence_level": 0.95
}

curl -X POST https://cortex.phaeth.com/v1.0/models/mdl-demand-v1/simulate \
  -H "Authorization: Bearer $LICENSE_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "entity": "SKU-4521",
    "baseline": {"price": 29.99, "promotion_active": false,
                 "competitor_price": 32.50, "weather_temp_c": 24.0},
    "scenarios": [
      {"name": "price_increase_10pct", "changes": {"price": 32.99}},
      {"name": "run_promotion", "changes": {"promotion_active": true}}
    ],
    "horizon": 7
  }'
# run_promotion → +24.7% units · revenue impact +$19,154 / 7d

from phaeth_cortex import CortexClient

client = CortexClient(api_key="your-api-key")

# Get a 7-day demand forecast
forecast = client.predict(
    model_id="mdl-demand-v1",
    inputs=[{
        "sku_id": "SKU-4521",
        "price": 29.99,
        "promotion_active": False,
        "day_of_week": 2,
        "competitor_price": 32.50,
        "weather_temp_c": 24.0,
    }],
    horizon=7,
)

print(forecast.predictions[0].forecast)
# [{"date": "2026-06-23", "value": 342, "lower": 295, "upper": 389}, ...]

Same API surface. Two ways to run it.

Both modes expose identical routes, SDKs, and the OpenAPI spec, the only difference is the base URL and where your data lives. Start on hosted SaaS, move to your VPC when compliance demands it, without rewriting a line.

Nine steps to a forecasting agent.

The path from a fresh API key to forecasts surfaced in your product. Each step maps to one or two endpoints above.

1. Register: obtain your license_token and account_id from /v1.0/auth/register.
2. Upload data: POST a dataset with a schema and a data URL (Parquet, CSV, or JSON).
3. Train: POST to /v1.0/models/train with target, horizon, and entity column.
4. Validate: confirm MAE / RMSE / MAPE / R² meet your business requirements.
5. Wire predict: register predict as a tool in your agent's calling framework.
6. Wire simulate: enable what-if reasoning so the agent can compare scenarios.
7. Stream data: append new rows daily or hourly to keep the model current.
8. Configure alerts: notify on accuracy degradation or detected drift.
9. Surface forecasts: display predictions with confidence intervals in your UI.

Pick your stack.

First-class SDKs for the two languages most agent platforms ship in, plus a machine-readable spec you can generate clients or MCP tools from.