Skip to main content
LLM API serves models from many labs under one API key, including Amazon Nova, Meta Llama, Mistral, Qwen, Google Gemini and Gemma, Anthropic Claude, and OpenAI GPT models. LLM API supports Chat Completions, Responses, and Anthropic Messages requests, and the Images API for its image models.
LLM API is a built-in provider. You can pay for many of its models with credits, and the rest need your own LLM API key. The Needs BYOK? column in the model catalog shows which is which.

​
Setup

1

Create an access key and add credits

Follow the quickstart to create an access key and purchase credits in app.ngrok.ai.
2

Store your LLM API key (optional)

Skip this step to pay with credits for the models that support it. To bring your own key, create an API key in your LLM API account. Then add a provider key and create an access key configuration with an llmapi routing rule that uses it.
3

Send requests

Send requests with your access key:
Many LLM API models are also served by their own labs. Add the llmapi: prefix, as in llmapi:google/gemma-4-31b-it, to keep the request on LLM API.

​
Which models work with credits

LLM API can send a request to a different upstream than the one the model names, and it bills the price of the upstream it picks. A model works with credits only when no upstream that LLM API might pick charges more than the catalog price. Models with long-prompt price tiers, separate cache-write prices, or image input priced apart from text need your own LLM API key. That includes every Claude model.

​
Request formats

  • Chat Completions and Responses: Most chat models accept Chat Completions only. The OpenAI models, xai/grok-4.7, meta/muse-wonder-1.0, and qwen/qwen3-coder-next also accept Responses.
  • Messages: LLM API serves the Anthropic Messages API for Claude models only.
  • Images: The image models, such as openai/gpt-image-2, z-ai/glm-image, and qwen/qwen-image-2.0, serve the Images API at /v1/images/generations. Some of them also serve /v1/images/edits.
The gateway removes these fields before it sends a request to LLM API, and lists them in the Ngrok-AIG-Removed-Params response header:
  • web_search_options and web_search, because LLM API runs and bills its own web search when either is present.
  • A top-level cache_control.
  • store and previous_response_id on Responses requests, because LLM API doesn’t store responses.

​
Available models

See the model catalog.

​
Next steps