Appearance

Models ​

Models are the AI engines your app or Agent can use.

SandBase lets you browse Models and APIs, inspect capabilities, copy a working request, and use the same capability inside Setup or Build Agent. Each model page documents the exact identifier, input fields, output shape, and endpoint to call. Browse the separate API Catalog for non-model capabilities such as search, scraping, data, and SaaS actions.

Where Models fit ​

PlaceHow Models are used
StoreBrowse Models and inspect pricing, IDs, and capabilities
SetupAdd selected Models to your AI tool
Build AgentPick the Model your Agent should use
API ReferenceCall Models directly from your app

Model API surfaces ​

Use the API surface that matches the model category:

Model categoryEndpointHow it works
LLMsPOST /v1/chat/completionsSend messages and receive a provider-compatible text response.
Image modelsPOST /v1/runSubmit an image generation task, then poll GET /v1/run/{id} for the result.
Video modelsPOST /v1/runSubmit an asynchronous video task, then poll GET /v1/run/{id} until it is terminal.
Audio and other capabilitiesPOST /v1/runUse the model-specific fields documented on its API reference page.
API Store capabilitiesGET/POST /v1/api/{vendor}/{upstream_path}Put the vendor-qualified capability name in the URL; a POST body can omit model.
Provider-native APIsProvider-specific endpointUse the native protocol and authentication shown in the Official Native API section.

The shared /v1/run endpoint is the standard SandBase interface for image and video models. Requests return an opaque run ID for asynchronous capabilities; keep polling the run endpoint for authoritative status and outputs. See Run a capability and Get a run result for the common request and response contract.

API Store capabilities are addressed by URL instead: use API Passthrough with /v1/api/{vendor}/{upstream_path}. This URL-derived route is separate from model generation through /v1/run.

Call an LLM ​

bash
curl https://api.sandbase.ai/v1/chat/completions \
  -H "Authorization: Bearer $SANDBASE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek/deepseek-v4-flash",
    "messages": [{"role": "user", "content": "Write a short customer brief."}]
  }'

Call an image or video model ​

Image and video models use the same asynchronous Run API. The model value is the exact SandBase model ID, while the remaining fields come from that model's reference page.

bash
# Submit an image or video task
curl -X POST https://api.sandbase.ai/v1/run \
  -H "Authorization: Bearer $SANDBASE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "<model-id>",
    "prompt": "A cinematic product shot at sunrise."
  }'

# Poll the returned id until status is completed or failed
curl https://api.sandbase.ai/v1/run/<run-id> \
  -H "Authorization: Bearer $SANDBASE_API_KEY"

For model-specific parameters, response fields, limits, and working examples, open the model's API reference from Supported Models.

Use a Model in an Agent ​

In Build Agent, the Model is one part of the Agent.

The Agent can also include APIs, Skills, instructions, and credentials.

Model capabilities ​

Different Models support different input and output types.

Examples:

  • text in / text out
  • image input
  • image generation
  • audio
  • video
  • embeddings
  • tool calling
  • structured output

Check capability details before using a Model in production.

Next steps ​