FAQ
How much history do I need?
At least the model's minimum context: 16 points per entity for hora-1.0. Points filled by cleaning count toward the length, but a long gap just before the forecast start, or a missing share of 35% or more, blocks the request (see Missing-data rule). See Models & Limits.
What is the maximum horizon h?
The request field allows 1 to 2048, but the model has its own maximum: 1,024 for hora-1.0. Going above it returns HORIZON_EXCEEDS_MODEL.
Is freq required?
Under the default preset it may be omitted and the service infers it; for irregular data or several entities, pass it explicitly. The raw preset requires it. Note that lowercase m means minutes; month end is ME or M, month start is MS. See Frequency strings.
What about missing or duplicate data?
The default preset de-duplicates, sorts and fills short gaps automatically, and writes every change to the cleaning report. If you are unsure what will happen to your data, dry-run with the free /v1/clean first. If you do not want your data modified, use the strict preset or clean_series: false.
What if my timestamps have no time zone?
They are parsed in options.timezone; without it they are read as UTC and the cleaning report says so. Timestamps in responses are always UTC. See Time zones.
Which requests are metered?
Only successful /v1/forecast calls, counted as "observations fed to the model + forecast points". Error responses and /v1/clean are never metered. See Metering, Quotas & Rate Limits.
What should I do on a 429?
Check X-RateLimit-Policy: for forecast-rpm and clean-rpm, retry after Retry-After seconds; forecast-daily-points only clears at the next UTC midnight.
Are covariates supported?
The API accepts them: extra numeric fields on series rows are past covariates, and future_covariates carries values over the forecast window. No model uses covariates today; each column is reported model_unsupported. The per-column report in model_input[].covariates is authoritative. See Covariate Support.
Are idempotency keys supported?
Yes, on POST /v1/forecast. Idempotency-Key is scoped to the API Key and kept for 24 hours; a retry with the same key and body returns the first stored result (with X-Idempotent-Replayed: true) and is not billed again. Upstream 5xx, timeouts and 429 are not stored, so you can retry them with the same key. See Authentication & Headers.
Can I call the API from a browser?
Not recommended. An API Key is an account credential: use it on the server only, and never put it in front-end code or commit it to a repository.
Can I use a sqx_test_* key?
No. The live API accepts sqx_live_* only and returns INVALID_API_KEY for test keys, and the docs site has no mock sandbox. To see results first, try the forecast workbench; to call the API, get an API Key on the SequenxAI Platform — the free tier is enough.
Which model does the API use?
SequenxAI Omni, model id hora-1.0 (alias hora), which is also used when model is omitted; any other id returns 422 MODEL_NOT_AVAILABLE. Its behavior can be adjusted with model_options, see Models & Limits. We make no accuracy claims; backtest on your own data before relying on the forecasts.