Direct model access

The LLM Gateway is in beta.

Direct model access exposes each provider API through a provider-specific gateway path. The gateway still handles authentication, provider secrets, policies, and tracing, but it does not translate the request and response into another provider’s API format. Direct model access is the path LangChain and Deep Agents use by default. For other clients, use it when you want a provider’s native request and response behavior. To call models across providers in one API format, use the standard endpoint instead.

Choose a provider path

Append a provider path to your regional gateway base URL, for example https://eu.gateway.smith.langchain.com/anthropic for direct Anthropic access in GCP EU:

Provider	Gateway path
Anthropic	`/anthropic`
AWS Bedrock	`/bedrock`
Azure Foundry	`/azure`
Baseten	`/baseten`
Fireworks	`/fireworks`
Google Gemini	`/gemini`
Gemini Enterprise Agent Platform	`/vertex`
OpenAI	`/openai`
ScaleDown	`/scaledown`
TypeSafe	`/typesafe`
xAI	`/xai`

Each path uses the provider key stored in the caller’s workspace. For secret names, see Provider authentication. Gateway Credits models use the standard endpoint rather than a provider-specific path. These hosted models require no provider secret of your own.

Configure provider SDKs

Set each provider SDK’s base URL to its direct gateway path and use your LangSmith API key as the provider API key:

export LANGSMITH_API_KEY="lsv2_..._....cbed3e"
export BASE_URL="https://gateway.smith.langchain.com"

export ANTHROPIC_BASE_URL="$BASE_URL/anthropic"
export OPENAI_BASE_URL="$BASE_URL/openai/v1"
export GOOGLE_GEMINI_BASE_URL="$BASE_URL/gemini"

export ANTHROPIC_API_KEY="$LANGSMITH_API_KEY"
export OPENAI_API_KEY="$LANGSMITH_API_KEY"
export GEMINI_API_KEY="$LANGSMITH_API_KEY"
export GOOGLE_API_KEY="$LANGSMITH_API_KEY"

The gateway resolves the actual provider key provider secrets, so the provider key does not need to be stored locally.

import os

from openai import OpenAI

client = OpenAI(
    base_url=os.environ["OPENAI_BASE_URL"],
    api_key=os.environ["LANGSMITH_API_KEY"],
)
response = client.chat.completions.create(
    model="gpt-5.4-mini",
    messages=[{"role": "user", "content": "ping"}],
)
print(response.choices[0].message.content)

import os

import anthropic

client = anthropic.Anthropic(
    base_url=os.environ["ANTHROPIC_BASE_URL"],
    api_key=os.environ["LANGSMITH_API_KEY"],
)
message = client.messages.create(
    model="claude-opus-5",
    max_tokens=1024,
    messages=[{"role": "user", "content": "ping"}],
)
print(message.content[0].text)

Direct paths use the provider’s native model name without a provider prefix.

Configure LangChain and Deep Agents

LangChain chat models and Deep Agents, including Deep Agents Code, reach these provider paths through the LANGSMITH_GATEWAY environment variable:

export LANGSMITH_API_KEY="lsv2_..."
export LANGSMITH_GATEWAY="true"

Deep Agents Code also requires LANGSMITH_GATEWAY_API_KEY. Set it to the same key:

export LANGSMITH_GATEWAY_API_KEY="$LANGSMITH_API_KEY"

LANGSMITH_GATEWAY support requires these minimum Python package versions: langchain-anthropic>=1.5.1, langchain-baseten>=0.2.3, langchain-fireworks>=1.5.1, langchain-google-genai>=4.3.2, and langchain-openai>=1.4.1.

Each supported integration appends its own provider path, so ChatAnthropic calls https://gateway.smith.langchain.com/anthropic and ChatOpenAI calls https://gateway.smith.langchain.com/openai/v1. To route requests through the standard endpoint instead, see Send your first LLM Gateway request.

Get started

Call the gateway

Models and reliability

Authentication

Observability

Governance

Choose a provider path

Configure provider SDKs

Configure LangChain and Deep Agents

See also

​Choose a provider path

​Configure provider SDKs

​Configure LangChain and Deep Agents

​See also

Choose a provider path

Configure provider SDKs

Configure LangChain and Deep Agents

See also