GloopAPI integration guide
Model selection
Discover models available to the current key.
Model IDs must come from the catalog available to your key. The anonymous model showcase is for browsing, not execution authorization or a guarantee of endpoint capabilities, pricing, or production enablement.
Text and protocols
Read the model catalog available to your key first. Lists and details use the OpenAI format. Call chat models through Chat Completions or Responses and check the required capabilities before sending a request.
Image inputs require a vision-capable model. Structured output and tool calling depend on the selected model; a listed ID does not guarantee support for every optional field.
Image and video capabilities
Persistent workflows should read their dedicated capability catalogs, check feature switches, input modes, sizes, and complete presets, then obtain quotes. The current persistent image baseline is Ark Seedream Pro/Lite and Qwen. Synchronous Venice text-to-image produces one image and does not support reference images; synchronous capabilities are not promises about the persistent catalog.
The video catalog returns complete executable presets. Use each preset as a unit; do not assemble resolution, duration, audio, and aspect ratio from different modes. Frame inputs must use managed files allowed by the model. Related image modes may be absent when public HTTPS signed inputs are not ready.
Versions and run records
Aliases identify public models and do not automatically inherit model authorization. New tasks use the routes and prices effective at that time; accepted tasks retain execution and billing evidence. Reconfirm if the catalog or pricing changes after quoting. Recover existing tasks using their original request key and identity.
If the capability catalog is empty, check permissions and feature switches first; do not probe by polling generation endpoints. See Error handling and Billing for error and charge boundaries.