What the API can do
Model catalog
GET /api/v1/models returns model purpose, fields, limits, and estimated cost so integrations do not rely on guesswork.
One generation request
POST /api/v1/generations uses the same shape across models: type, model, and input.
Assets and references
Upload or import media, receive an asset URI, and pass it into model fields that accept references.
IDEMPOTENCY_KEY="${IDEMPOTENCY_KEY:-$(uuidgen)}"
curl -X POST https://api.givon.ai/api/v1/generations \
-H "Authorization: Bearer $GIVON_API_KEY" \
-H "Idempotency-Key: $IDEMPOTENCY_KEY" \
-H "Content-Type: application/json" \
-d '{"type":"image","model":"gpt-image-2.5-flare","input":{"prompt":"a cat in glasses reading a book"}}'How the developer workflow works
Start with the catalog endpoint. It gives the available public image and video models, their required fields, supported options, and estimated token use. That catalog is the integration contract, so client code can stay aligned as model options evolve.
For media references, create an owner-media asset first. The returned asset URI can then be used in model inputs for reference images, starting frames, source videos, or other supported fields.
Why MCP matters for AI agents
The MCP server lets agent tools call Givon AI with the same account key and the same model definitions as the REST API. This is useful when the agent decides which model fits the task, checks the schema, and assembles the request itself.
Use it for internal content pipelines, creative assistants, marketplace image automation, and workflows where a human should review the result rather than manually repeat the generation setup every time.
API and MCP questions
Where do I create an API key?
Keys are managed in the Givon AI account, in the developer integrations area.
Do API generations use the same balance?
Yes. API and MCP generations use your Givon AI token balance.
