Trace OpenAI-compatible providers

Trace LLM calls from any OpenAI-compatible provider to LangSmith.

Many LLM providers accept requests in the same format as the OpenAI API. To trace calls from these providers to LangSmith, construct an OpenAI client pointed at the provider’s base URL, then wrap it with wrap_openai / wrapOpenAI.

Use wrap_openai / wrapOpenAI for direct API calls. Use @traceable when you need to trace application logic around the call or set metadata per invocation.

	wrap_openai / wrapOpenAI	@traceable / traceable
Token tracking	Automatic	Requires run_type="llm"
Run type	LLM (set automatically)	Chain by default
Traces	The API call	The function wrapping it
Metadata	Client-level only (Python); client-level or per-call (TypeScript)	Per-call via langsmith_extra

To trace OpenAI directly, refer to Trace OpenAI applications.

Setup

Python

pip install langsmith openai

npm install langsmith openai

export LANGSMITH_API_KEY=<your-api-key>
export LANGSMITH_TRACING=true

Trace API calls

Python

import os

import openai
from langsmith import wrappers

client = wrappers.wrap_openai(
    openai.OpenAI(
        base_url="https://<provider-base-url>/v1",
        api_key=os.environ["PROVIDER_API_KEY"],
    )
)

completion = client.chat.completions.create(
    model="<provider-model-name>",
    messages=[{"role": "user", "content": "Hello!"}],
)
print(completion.choices[0].message.content)

import OpenAI from "openai";
import { wrapOpenAI } from "langsmith/wrappers/openai";

const client = wrapOpenAI(
  new OpenAI({
    baseURL: "https://<provider-base-url>/v1",
    apiKey: process.env.PROVIDER_API_KEY!,
  })
);

const completion = await client.chat.completions.create({
  model: "<provider-model-name>",
  messages: [{ role: "user", content: "Hello!" }],
});
console.log(completion.choices[0].message.content);

Add metadata

Python

Pass tracing_extra when wrapping the client. The metadata applies to all calls made with that client.

import os

import openai

from langsmith import wrappers

client = wrappers.wrap_openai(
    openai.OpenAI(
        base_url="https://<provider-base-url>/v1",
        api_key=os.environ["PROVIDER_API_KEY"],
    ),
    tracing_extra={"metadata": {"environment": "production"}},
)

TypeScript

Pass options as the second argument to wrapOpenAI for client-level metadata, or pass langsmithExtra per call.

import OpenAI from "openai";
import { wrapOpenAI } from "langsmith/wrappers/openai";

const client = wrapOpenAI(
  new OpenAI({
    baseURL: "https://<provider-base-url>/v1",
    apiKey: process.env.PROVIDER_API_KEY!,
  }),
  { metadata: { environment: "production" } }
);

// Per-call metadata
const completion = await client.chat.completions.create(
  {
    model: "<provider-model-name>",
    messages: [{ role: "user", content: "Hello!" }],
  },
  { langsmithExtra: { metadata: { request_id: "abc123" } } }
);

Related guides

Some providers have dedicated setup guides that use @traceable or a native callback. These approaches trace at the function level rather than wrapping the client directly, or integrate with the provider’s own SDK and routing layer.

• DeepSeek: OpenAI-compatible API; guide uses @traceable with custom provider metadata
• LiteLLM: proxy that exposes an OpenAI-compatible endpoint; guide covers @traceable and LiteLLM’s built-in LangSmith callback

---

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

Edit this page on GitHub or file an issue.