Errors & limits

Handle error responses, retry safely, and understand service limits and usage.

Error responses

Error response

{
  "error": {
    "type": "api_error",
    "code": "data_unavailable",
    "message": "The requested market window is unavailable.",
    "httpStatus": 422,
    "requestId": "…"
  },
  "requestId": "…"
}

400 / 422invalid input or coverage: Correct the field identified by param, or supply a supported market window.
401 / 403authentication or scope: Check the key, expiration, workspace and required scopes.
404not found: The job/image is not in this workspace, or its retention period ended.
409idempotency conflict: The same Idempotency-Key was used with a different request.
429rate or queue limit: Wait as directed by Retry-After before submitting more work.
502 / 503upstream or renderer: Inspect the error code; retry transient service failures with bounded backoff.

Retry without duplicating work

Set a unique Idempotency-Key on each new logical POST. A retry with the same key and body returns the original resource. A different body under that key returns 409. Results expire after retention; idempotency tombstones remain to prevent accidental replay; never blindly retry a completed analysis with a fresh key.

Current service limits

Candles2–2000: One market and timeframe per job. Historical public data requests are bounded and may fail if upstream coverage cannot be satisfied.
ImagePNG: 640–2560px wide; 360–1600px high. Default 1440 × 900.
Commands100 per request: Validated typed chart commands only.
Queue20 pending/workspace: Two active jobs per worker by default; configurable by the server owner.
Progress streams8/workspace · 64/process: An excess stream returns 429 stream_limit_exceeded with Retry-After: 5. Reconnect with Last-Event-ID rather than opening duplicate streams.
Retention7 days: Download results needed for longer research or audit retention.

Key rate and daily quota are set at provisioning. Inspect response rate headers and GET /v1/capabilities for server capabilities. There are no published paid API tiers or automatic overage charges in this preview.

Usage and billing

Job usage reports managed brain invocations, successful renders, render attempts and model tokens when available. One brain invocation may make multiple provider requests. Failed or cancelled jobs retain known work; missing token usage is unknown, not necessarily zero.

With audit:read, the existing workspace usage endpoint returns attributed request units, token totals and configured cost estimates. Those units are rate-limit weights, not purchased credits. Zero estimated cost can mean model rates are unconfigured. Card charging, commercial allowances and API overages are not enabled by this preview.