The process of model calling with Vercel AI Gateway consists of three key components:
- Authentication Configuration: Set the proprietary API_KEY obtained on the console as an environment variable
AI_GATEWAY_API_KEYReplacement of the original vendor key - endpoint: When initializing the client, the
base_urlThe parameter is specified as the Vercel gateway addresshttps://ai-gateway.vercel.sh/v1 - Model Selection: Adoption of a standardized format
供应商/模型名称Specify the target model, e.g.anthropic/claude-sonnet-4
As an example, the core modifications to the Python code include:
- Redirection of base_url required for OpenAI client initialization
- The model parameter in the request needs to be in vendor prefix format
- Authentication uses Vercel gateway keys instead of stock keys
This design allows model switching by modifying only the model parameter strings, which dramatically improves development flexibility.
This answer comes from the articleVercel AI Gateway: a gateway to manage and optimize AI application requestsThe
May not be reproduced without permission:AI productivity tools " How to call AI models from different vendors via Vercel AI Gateway? What core parameters need to be modified?
English 
简体中文 
日本語 
Deutsch 
Português do Brasil