Skip to Content
AI GatewayFrameworks

Frameworks

Keep your framework—override the OpenAI-compatible (or Anthropic) base URL and use provider/model ids where the client speaks Chat Completions.

Unified vs passthrough for frameworks

Framework patternRecommended surface
LangChain / Vercel AI / CrewAI / ADK with provider/modelUnified /gateway/v1 (examples below)
OpenAI Agents SDK / Agno OpenAIChat / ADK→OpenAI onlyOpenAI passthrough /gateway/openai/v1 — Passthrough
Claude Agent SDK / Agno ClaudeAnthropic passthrough /gateway/anthropic
Cursor / Claude Code IDEsCursor / Claude Code

Decision guide: Choose your surface.

LangChain

from langchain_openai import ChatOpenAI
import os
 
llm = ChatOpenAI(
    model="anthropic/claude-sonnet-4-6",
    api_key=os.environ["EXEMPLAR_API_KEY"],
    base_url=os.environ.get(
        "EXEMPLAR_GATEWAY_URL",
        "https://production-api.exemplar.dev/gateway/v1",
    ),
)
llm.invoke("Summarize open incidents")
import { ChatOpenAI } from "@langchain/openai";
 
const llm = new ChatOpenAI({
  model: "anthropic/claude-sonnet-4-6",
  apiKey: process.env.EXEMPLAR_API_KEY!,
  configuration: {
    baseURL:
      process.env.EXEMPLAR_GATEWAY_URL ??
      "https://production-api.exemplar.dev/gateway/v1",
  },
});
 
await llm.invoke("Summarize open incidents");

Vercel AI SDK

import { createOpenAI } from "@ai-sdk/openai";
import { generateText } from "ai";
 
const gateway = createOpenAI({
  apiKey: process.env.EXEMPLAR_API_KEY!,
  baseURL:
    process.env.EXEMPLAR_GATEWAY_URL ??
    "https://production-api.exemplar.dev/gateway/v1",
});
 
const { text } = await generateText({
  model: gateway("openai/gpt-4o-mini"),
  prompt: "One-line SRE tip",
});

Claude Agent SDK

Point the Agent SDK at the Anthropic passthrough (/gateway/anthropic). The SDK speaks Anthropic Messages; set ANTHROPIC_BASE_URL / ANTHROPIC_API_KEY (use your eis_* key).

import { query } from "@anthropic-ai/claude-agent-sdk";
 
for await (const message of query({
  prompt: "Summarize open incidents",
  options: {
    model: "claude-sonnet-4-6",
    env: {
      ...process.env,
      ANTHROPIC_BASE_URL: "https://production-api.exemplar.dev/gateway/anthropic",
      ANTHROPIC_API_KEY: process.env.EXEMPLAR_API_KEY!,
    },
  },
})) {
  console.log(message);
}
import os
from claude_agent_sdk import query, ClaudeAgentOptions
 
options = ClaudeAgentOptions(
    model="claude-sonnet-4-6",
    env={
        "ANTHROPIC_BASE_URL": "https://production-api.exemplar.dev/gateway/anthropic",
        "ANTHROPIC_API_KEY": os.environ["EXEMPLAR_API_KEY"],
    },
)
 
async for message in query(prompt="Summarize open incidents", options=options):
    print(message)

TypeScript options.env replaces the process environment—spread process.env as shown. Python ClaudeAgentOptions(env=…) merges on top of the inherited environment. Enable the anthropic provider under Management → Providers.

Google ADK

Use ADK’s LiteLLM / OpenAI-compatible shim against /gateway/v1. LiteLLM expects an openai/ prefix before the gateway model id.

from google.adk.models.lite_llm import LiteLlm
import os
 
llm = LiteLlm(
    model="openai/gemini/gemini-2.5-flash",  # LiteLLM + gateway provider/model
    api_base=os.environ.get(
        "EXEMPLAR_GATEWAY_URL",
        "https://production-api.exemplar.dev/gateway/v1",
    ),
    api_key=os.environ["EXEMPLAR_API_KEY"],
)
# Attach llm to your ADK LlmAgent / root agent as usual.
import { createOpenAI } from "@ai-sdk/openai";
 
// JS/TS ADK stacks that take an OpenAI-compatible provider:
const gateway = createOpenAI({
  apiKey: process.env.EXEMPLAR_API_KEY!,
  baseURL:
    process.env.EXEMPLAR_GATEWAY_URL ??
    "https://production-api.exemplar.dev/gateway/v1",
});
// model: gateway("gemini/gemini-2.5-flash") or gateway("openai/gpt-4o-mini")

OpenAI Agents SDK

import { Agent, run } from "@openai/agents";
 
process.env.OPENAI_BASE_URL =
  process.env.EXEMPLAR_GATEWAY_URL ??
  "https://production-api.exemplar.dev/gateway/v1";
process.env.OPENAI_API_KEY = process.env.EXEMPLAR_API_KEY!;
 
const agent = new Agent({
  name: "Platform copilot",
  model: "openai/gpt-4o-mini",
  instructions: "You are an SRE assistant.",
});
await run(agent, "Summarize open incidents");
import os
from agents import Agent, Runner
 
os.environ["OPENAI_BASE_URL"] = os.environ.get(
    "EXEMPLAR_GATEWAY_URL",
    "https://production-api.exemplar.dev/gateway/v1",
)
os.environ["OPENAI_API_KEY"] = os.environ["EXEMPLAR_API_KEY"]
 
agent = Agent(
    name="Platform copilot",
    model="openai/gpt-4o-mini",
    instructions="You are an SRE assistant.",
)
Runner.run(agent, "Summarize open incidents")

CrewAI

from crewai import LLM, Agent, Crew, Task
import os
 
llm = LLM(
    model="openai/gpt-4o-mini",
    base_url=os.environ.get(
        "EXEMPLAR_GATEWAY_URL",
        "https://production-api.exemplar.dev/gateway/v1",
    ),
    api_key=os.environ["EXEMPLAR_API_KEY"],
)
researcher = Agent(role="SRE Analyst", goal="Triage incidents", llm=llm)
crew = Crew(
    agents=[researcher],
    tasks=[Task(description="Summarize open alerts", expected_output="A short triage note")],
)
crew.kickoff()
Last updated on
Provider passthroughCursor IDE