Skip to main content

LangChain.js

LangChain.js ChatOpenAI works against Ferro Labs AI Gateway as a drop-in by setting configuration.baseURL. A first-party @ferro-labs-ai/sdk/langchain sub-export is planned.

Drop-in (today)​

pnpm add @langchain/openai @langchain/core
import { ChatOpenAI } from "@langchain/openai";
import { HumanMessage } from "@langchain/core/messages";

const chat = new ChatOpenAI({
  model: "claude-3-5-sonnet-20241022",      // routed to Anthropic by Ferro
  apiKey: process.env.FERRO_API_KEY,
  configuration: { baseURL: process.env.FERRO_BASE_URL + "/v1" },
});

const response = await chat.invoke([new HumanMessage("Hello from LangChain.js")]);
console.log(response.content);

Read the trace_id​

The x-trace-id response header is the canonical Ferro trace ID. LangChain.js exposes raw response metadata via the additional_kwargs and response_metadata fields, but the cleanest path today is to read it directly from your fetch wrapper or via @ferro-labs-ai/sdk for typed access.

Verify​

curl -i http://localhost:8080/v1/chat/completions \
  -H "Authorization: Bearer $FERRO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-4o-mini","messages":[{"role":"user","content":"hi"}]}' \
  | grep -i '^x-trace-id'

First-party adapter (planned)​

@ferro-labs-ai/sdk/langchain will ship a FerroChatModel mirroring the Python adapter — typed Ferro extras (routeTag, templateId, templateVariables), traceId on response_metadata, native LangGraph.js compatibility. The npm scope @ferro-labs-ai is already reserved.

Track progress: Ferro AI Gateway roadmap.

Runnable example​

ai-gateway-cookbook/typescript/02-langchainjs-streaming-agent is planned alongside the sub-export release.

See also​