Skip to content

stampede ai jobs create

Start a job that generates a scenario (asynchronous)

Start a generation job on the server. Give at least one input: a description, an OpenAPI document, a HAR recording, an access log or .proto files. Recorded traffic is redacted before it reaches the provider. With –target, every journey is dry-run against that target and repaired; with –scenario, the proposal is compared with that scenario and approving it saves a new version. Prints the job id, or with –wait follows the job and shows the result.

stampede ai jobs create [flags]
stampede ai jobs create --describe "people browse items, add one to the cart and check out" \
--from-openapi openapi.yaml --target staging --wait
stampede ai jobs create --from-har session.har --target staging --scenario checkout
--describe string plain-language description of your users and what they do
--from-har string HAR recording of real use
--from-log string web server access log, used to estimate the journey mix
--from-openapi string OpenAPI 3.x document (YAML or JSON)
-h, --help help for create
--json print JSON for scripting
--max-repairs int repair rounds before a failing journey is flagged for a human, 0 to 3 (default 3)
--no-dry-run skip the dry run and only check the scenario statically
--proto stringArray .proto file whose services become grpc steps (repeatable; imports name files by base name)
--provider string AI provider name or id (default: the only one, or the one named default)
--scenario string existing scenario to compare the proposal with (approving saves a new version)
--target string target to dry-run against (name, base URL or id); without it the scenario is only checked statically
--wait wait for the job to finish and show it
--project string project name, slug or id (default: the default or only project)