cost line is what you were actually charged, and what that works out to
per result. The held line is what was reserved before the call ran. The
difference was released back to your balance.
Options
Plus the global options.
-i and -f are
mutually exclusive, and passing both is a usage error before anything is sent.
Input from a file or stdin
Long or generated inputs are easier from a file:Sending a local image
The video tools animate a starting image.--image is the whole file intake in
one flag:
.jpg that is really a PNG still works. JPEG, PNG and WEBP,
up to 10 MB.
The upload happens before the run, so a file the gateway will not take costs
nothing: no hold, no run row. --image supplies the image on its own, so
setting image_handle, image_urls or image_base64 in the input as well is a
usage error before anything is sent.
The two requests behind that one flag exist for agents rather than for you.
Tool arguments pass through a model’s context window, so an image sent inline
as base64 costs a million tokens and gets truncated. Uploads
explains the flow and the 10 and 30 minute clocks on an uploaded file.
Slow calls
Some calls take minutes. If a run does not finish inside the gateway’s synchronous window, you get the run id back instead of the rows:--wait to block until it finishes instead. The CLI polls with exponential
backoff up to --timeout seconds:
Ctrl-C during
--wait stops the waiting, not the run. The run keeps going and
you can pick it up later with goro runs get -r <runId>.4 and tells
you how to keep waiting. The run itself is unaffected.
Saving the output
Human output prints the first three rows and then points you at the full result. To keep everything:When it does not work
A 402 prints what it needs and where to fix it:
Next: runs
Poll a run, and see what you have spent.