SequenxAI
ZHContact

SequenxAI Platform · Time-series forecasting service

One API.Cleaned, then forecast.Data that can stay in China.

One API for time-series forecasting: the general model works out of the box, and enterprise engagements add domain tuning by industry. Sensitive data stays onshore, with on-premises deployment per project; cleaning handles dropouts, duplicates and off-grid timestamps automatically, flags outliers without changing the original data, and returns a quality report. The platform is in beta.

Try the SequenxAI PlatformRead the architecture deep-dive

One request · forecast and cleaning report Illustrative data

Field names from the public API · values are placeholders

Request
POST /v1/forecast
{
  "model": "hora-1.0",
  "series": [
    {"unique_id": "meter-001",
     "ds": "…T00:00:00Z", "y": …},
    …
  ],
  "h": 24,
  "freq": "1h",
  "level": [80],
  "clean_series": true
}
Response
200 OK
{
  "model_used": "hora-1.0",
  "forecast": [
    {"ds": "…", "y_hat": …,
     "intervals": {
       "80": {"lo": …, "hi": …}}},
    …
  ],
  "data_quality": {
    "status": "cleaned",
    "imputed_points": …,
    "notes": ["…"]
  }
}
y_hat and the 80% interval per step (illustrative)

Illustrative data, not a live forecast.P10–P90

Why the SequenxAI Platform

Compliance, cleaning and one interface,without assembling them yourself.

Overseas APIs / self-hosted open weights

  • Compliance: overseas APIs send business data out of the country
  • Residency: data sits on overseas nodes, or you build onshore and private environments yourself
  • Cleaning: models do not clean data — gaps, duplicates and off-grid timestamps are yours to fix
  • Interfaces: every model library has its own input format, so switching and audits need per-model work

SequenxAI Platform

  • Compliance: mainland endpoints, with on-premises deployment per project
  • Robustness: cleaning pipeline + auditable data-quality report; hard fail when unsafe
  • One contract: a single long-table format, the same from the general model to industry models
  • Transparent: quantile intervals and a quality report with every forecast

Input · Output

Designed for Real-World Sequence Specs

Accepts

  • Long-table time series (one or many entities) at one regular frequency per request (seconds to years); slightly off-grid timestamps are snapped to the grid by cleaning — not text, images, or video
  • Covariates: the API accepts and reports them per column; current models do not use them yet (planned)
  • REST ingestion today (POST /v1/forecast); batch, streaming and gRPC are planned, not yet available

Produces

  • Point forecast + quantile intervals (e.g. P10–P90; the model natively covers P1–P99)
  • Probabilistic scenario fan-charts and sampled paths (planned, not yet available)
  • Anomaly likelihood scores per timestep (planned, not yet available)
  • Structured decision-support signals (planned, not yet available)

Scope & Boundaries

What the SequenxAI Platform Is — and Isn’t

The SequenxAI Platform is:

  • A time-series forecasting service whose inputs are time series, with the general model and domain tuning behind one API
  • An enterprise time-series AI engine for projects in energy, transportation and other industries
  • Focused on CN residency and auditable cleaning
  • Delivery scope agreed in writing with each customer, per scenario

The SequenxAI Platform is not:

  • A general-purpose LLM or text / image / video generation model
  • An accuracy leader claimed without public same-scenario benchmarks
  • A generally available service: the forecast workbench and API Key management are both in beta today
  • A replacement for domain expert judgment

Get connected

Ready to integrate?Start with the developer docs

The developer docs cover fields, error semantics and call examples; to see how a request flows inside the platform, read the architecture deep-dive.

Open developer docsRead the architecture deep-dive