Run a task
Record a request to run a task, and answer with it PENDING.
region may be omitted for the organization’s default; whichever it was is
stored on the run, since the default can change afterwards.
Platform scripts capture caller inputs plus publisher-fixed dependencies and Spark configs, then dispatch in the region as the caller, without a trigger or window. Scripts allow only args, runtime, version, timeoutSeconds and disk sizes; publisher settings cannot be overridden through this API. Union runs retain their existing launch-intent workload path. Direct execution does not require the task-scheduling feature flag.
Repeated POSTs create new runs. There is no HTTP creation idempotency or automatic stale-launch recovery for a crash after commit but before enqueue; cancel clears a stranded PENDING script run. Results remain null.
201: the run, PENDING. 400: the task runs nowhere, or not in this region, or the inputs do not satisfy its schema, or the organization is not set up to run in the region, or the region has no Temporal namespace configured; an error naming every problem. A submission whose inputs do not satisfy the schema is also kept in run history as REJECTED, with those problems and its inputs (bounded and sanitized, as stored_inputs describes); nothing is launched or billed. 404: no task of this name the caller’s organization may see — the same answer a read gives, so asking to run reveals no more than reading. 503: the run was recorded but its launch could not be dispatched; the run is marked FAILED and the request may be retried. For platform scripts, a retry creates a new run; the failed run remains in history as a PLATFORM failure.
Authorizations
Bearer authentication header of the form Bearer <token>, where <token> is your auth token.
Path Parameters
The name of the task.
1 - 255^[a-zA-Z0-9_\-\.]+$Body
inputs must match the task's inputSchema from GET /tasks/{taskName}.
The task's examples show sample inputs to start from.
e.g., for POST /tasks/predict_mosaic_geometries_recipe/runs:
{
"region": "aws-us-west-2",
"inputs": {
"aoi_uri": "s3://wherobots-examples/rasterflow/aois/collegepark.parquet",
"start": "2023-01-01T00:00:00Z",
"end": "2024-01-01T00:00:00Z",
"model_recipe": "sam3-text-geometry",
"text_prompt": ["roofs"],
"confidence_threshold": 0.3,
"target_crs": "EPSG:3857"
},
"name": "College Park roofs, 2023"
}
Response
Successful Response
e.g. { "taskRunId": "npqy9q19q4wlm3", "taskName": "predict_mosaic_geometries_recipe", "state": "SUBMITTED", "region": "aws-us-west-2", "submittedAt": "2026-06-04T18:00:00Z" }
Current mutable Task display metadata, not historical identity.
PENDING, SUBMITTED, RUNNING, SUCCEEDED, FAILED, TIMED_OUT, CANCELLING, CANCELLED, UNKNOWN, SKIPPED, MISSED, REJECTED Concrete task polymorphic identities.
uniontask, scripttask Origin of a run, independent of its backend.
manual, schedule, rerun The task run this one reran, when runSource is rerun. Null otherwise.
What batchTime means: tick for a nominal schedule time, gap for a journal boundary such as a pause or schedule edit. Null when batchTime is null.
tick, gap Journal outcomes that deliberately launch no workload.
cancelled, overlap, epoch_gap, inputs_rejected Terminal dispatch and execution failure categories.
QUOTA_REJECTED, CONFIG_REJECTED, USER_CODE, PLATFORM, TIMEOUT Recorded execution identity; absent before launch, never inferred from Task Run ID.
How long the backend executed the run, from its start to its end (or to now, while running). 0 if it never started there. A cached run executes nothing, so this is the moment it took to return the stored result, not how long the result took to produce.

