TIR GenAI API

TIR GenAI API gives you ready-to-use inference endpoints for AI models, so you can call a model straight away and leave the serving infrastructure to E2E. Open a model, try it in the Playground, then call it from your own code.

Where to find it

In TIR, open your project. In the side menu’s PRODUCTS list, select GenAI API. The page that opens, Models, lists every model you can use.

The models on offer

Every model on GenAI API belongs to one task category: Text-to-Text, Image-to-Text, Image-to-Image, Object Detection, Speech-to-Text, Text-to-Image, Text-to-Speech, Multimodal, Text Embedding, and Rerank. Each model has a display name, author, category, and a short description, and you can search by name, author, or type.

What you can do

• Use the Playground to try a model and test prompts before you write any code.
• Run bulk prompts against a Text-to-Text model from an uploaded CSV file, and track the job in Async History.
• Call the API with the OpenAI SDK for a text or embedding model, using your existing OpenAI code with a different base URL.
• Call the API with REST for any model, with a plain HTTP request.
• Call the API with the TIR SDK for the small set of models it supports, such as Whisper Large V3, when the model’s name goes in the request body.
• Generate vector embeddings and store them in a TIR Vector Database.
• Build a RAG pipeline that answers questions from your own documents with a self-hosted model and the GenAI API’s embeddings.

For how a calling method matches a model, see Models and calling methods. For what a call costs, see Billing and usage.

Where to start

In Get started with TIR GenAI API, you try a model in the Playground and make your first API call. For help with a message, Troubleshooting TIR GenAI API has the fixes, and the TIR GenAI API FAQ has short answers.