What is an integration?

Overview

An integration endpoint lets an existing provider SDK call ModRouter without using ModRouter’s native /v1 request shape directly. The integration layer parses the SDK’s request format, converts it into ModRouter’s internal schema, runs the normal ModRouter pipeline, and converts the response back to the shape the SDK expects. Set the gateway root URL from the environment and append the integration route in code or config:

export MODROUTER_BASE_URL="https://api.modrouter.online"
export MODROUTER_API_KEY="<your-modrouter-virtual-key>"

MODROUTER_BASE_URL should not include /openai, /anthropic, /genai, or /v1; each integration page appends the route it needs. This is different from passthrough:

Mode	What ModRouter does	Use when
Integration	Converts request and response shapes	The endpoint is implemented by ModRouter and you want normal gateway behavior.
Passthrough	Forwards provider-native path and body through ModRouter’s provider pipeline	You need a provider endpoint that is not covered by a conversion route.

Endpoint Families

Integration	Base URL	Main request shape
OpenAI SDK	`/openai`	OpenAI API, including `/v1/chat/completions`, `/v1/responses`, files, batches, containers, images, audio, and selected realtime routes.
Anthropic SDK	`/anthropic`	Anthropic Messages, completions, model listing, count tokens, files, and message batches.
Google GenAI SDK	`/genai`	Gemini / GenAI model, file, batch, cached content, operation, and rank routes.
AWS Bedrock SDK	`/bedrock`	Bedrock Converse, Invoke, CountTokens, rerank, batch jobs, and S3-compatible file routes under `/bedrock/files`.
LiteLLM	`/litellm`	A combined adapter exposing OpenAI, Anthropic, GenAI, Bedrock, and Cohere-style routes.
LangChain	`/langchain`	A combined adapter for LangChain provider classes.
PydanticAI	`/pydanticai`	A combined adapter for PydanticAI provider classes.
Cohere	`/cohere`	Cohere v2 chat, embed, rerank, and v1 tokenize routes.
Cursor	`/cursor`	Cursor-specific compatibility for Cursor’s hybrid chat payloads, plus selected provider routes.

The plain ModRouter OpenAI-compatible API remains available at /v1.

Model Routing

Use provider/model model IDs when you want ModRouter to route explicitly:

response = client.chat.completions.create(
    model="anthropic/claude-sonnet-4-5",
    messages=[{"role": "user", "content": "Hello"}],
)

Unprefixed model names are interpreted by the route’s default provider when possible. Prefixed names are clearer in shared examples.

Authentication

Most SDKs require an API key locally even when ModRouter manages upstream provider keys. Use a ModRouter virtual key when governance is enabled. Otherwise, use a placeholder value accepted by the SDK. Common request auth options accepted by ModRouter include:

Authorization: Bearer <virtual-key>
X-Api-Key: <virtual-key>
x-bf-vk: <virtual-key>

List Models

ModRouter exposes model listing through the native /v1/models endpoint and through integration-specific routes such as /openai/v1/models, /anthropic/v1/models, and /genai/v1beta/models. For provider-specific model listing on the native endpoint, use:

curl "https://api.modrouter.online/v1/models?provider=openai"

Some integration list routes also accept ModRouter-specific filtering headers, but the native query form is the clearest option in docs.

​Overview

​Endpoint Families

​Model Routing

​Authentication

​List Models

Overview

Endpoint Families

Model Routing

Authentication

List Models