What is built: the workflow registry
A workflow is a named, versioned record that belongs to the workspace. It carries:- a name, up to 80 characters, and an optional description up to four kilobytes, which is stored encrypted;
- a declared type, one of
extraction,summarization,classification,coding,question_answering,content_generation,agent_workfloworunknown, or none; - an optional metric definition: how you will judge a run of this workflow succeeded, as an application assertion or a human acceptance, with a version so the definition can change without losing history;
- an archived flag. Archiving clears the description and does not restore it on unarchive.
What is still to come
In the order the work is planned:- Attribution. A request header naming the workflow it belongs to, read on the way through and never forwarded to a provider. A label never changes which model serves the request.
- Summaries. Request counts and charges rolled up per workflow, joined by request identity to the same records Activity shows.
- The usage screen. The roll-up beside the existing by-key and by-model views.
- Tasks. Bounded units of work inside a workflow, with their own scoped credentials.
- Deletion and retention. Removing a workflow’s observations on your own schedule.
Next
Monitor usage
The usage screen as it stands today, by key and by model.
Management API
Where the workflow endpoints live.