Unified endpoint - Docs by LangChain

Beta: The LLM Gateway is in beta.

The unified endpoint exposes the providers you’ve configured on the LLM Gateway behind a single base URL at the gateway root. Instead of pointing each client at a provider-specific path like /openai or /anthropic, you list models and call them from one place, addressing each model with a provider-prefixed ID such as openai/gpt-4o-mini, fireworks/accounts/fireworks/models/glm-5p2, or anthropic/claude-sonnet-4-6. This page shows you how to:

Set up the prerequisites for the unified endpoint.
List available models across configured providers.
Call a model with a provider-prefixed ID.
Understand how requests are translated between the Responses API and each provider’s native API.

Base URL

Point any OpenAI-compatible client at:

https://gateway.smith.langchain.com/v1

Authenticate with your LangSmith API key, passed by your clients as the provider API key.For regional base URLs, see Regional gateways.

Prerequisites

Before using the unified endpoint, confirm that:

Your Organization admin has added Provider Secrets for the providers you want to call. The unified endpoint resolves upstream credentials from the same workspace secrets as the provider-specific gateway paths. Refer to Admin setup.
You have a workspace-scoped LangSmith API key attached to a role with gateway:invoke and workspaces:read permissions.

List available models

Call GET /v1/models to list the models available to your workspace across providers:

curl https://gateway.smith.langchain.com/v1/models \
    -H "Authorization: Bearer $LANGSMITH_API_KEY"

The gateway aggregates each provider’s model catalog into a single OpenAI-compatible list. Every model ID is prefixed with the provider name in the form <provider>/<model>. The prefixed ID is what you pass as model when making a call:

{
  "object": "list",
  "data": [
    {
      "id": "openai/gpt-4o-mini",
      "object": "model"
    },
    {
      "id": "fireworks/accounts/fireworks/models/glm-5p2",
      "object": "model"
    },
    ...
  ]
}

Your response contains one entry per model from every provider reachable for your workspace. A provider whose Provider Secret isn’t configured is skipped rather than failing the request.

Call a model

Call models with POST /v1/responses, the OpenAI Responses API shape. Set model to the full provider-prefixed ID:

curl https://gateway.smith.langchain.com/v1/responses \
    -H "Authorization: Bearer $LANGSMITH_API_KEY" \
    -H "Content-Type: application/json" \
    -d '{"model":"openai/gpt-4o-mini","input":"ping"}'

curl https://gateway.smith.langchain.com/v1/responses \
    -H "Authorization: Bearer $LANGSMITH_API_KEY" \
    -H "Content-Type: application/json" \
    -d '{"model":"fireworks/accounts/fireworks/models/glm-5p2","input":"ping"}'

curl https://gateway.smith.langchain.com/v1/responses \
    -H "Authorization: Bearer $LANGSMITH_API_KEY" \
    -H "Content-Type: application/json" \
    -d '{"model":"anthropic/claude-sonnet-4-6","input":"ping"}'

You must ensure that the model field is provider-prefixed. A bare model name such as gpt-4o-mini returns a 400 error

The unified endpoint accepts Responses API requests.You can still reach your Anthropic models from the unified endpoint: call /v1/responses with an anthropic/-prefixed model ID and the gateway translates the request for you, as described below.

How requests are translated

The gateway translates between the Responses API and each provider’s native API in both directions. If the provider natively speaks the Responses API, as OpenAI and Fireworks do, your request and its response pass through unchanged. Otherwise, the gateway translates your request into the provider’s native format (for Anthropic models, the Messages API) and translates the response back into the Responses API format, including streaming responses. Translated or not, every call goes through the provider’s regular gateway path, so it resolves the same Provider Secrets, evaluates the same spend and redaction policies, and is traced like any other gateway call.

Next steps

Traces, Engine, and access control: where unified endpoint traces land and who can see them.
Model fallbacks: retry against backup models when a provider fails.
Custom model providers: route to your own OpenAI-compatible endpoints.
Spend policies: set cost limits on unified endpoint traffic.

Connect these docs to Claude, VSCode, and more via MCP for real-time answers.

Edit this page on GitHub or file an issue.

Base URL

​Prerequisites

​List available models

​Call a model

​How requests are translated

​Next steps

Prerequisites

List available models

Call a model

How requests are translated

Next steps