Create an LLM Monitoring Application¶
- Go to Agent Monitoring > Application List;
- Click Create;
- Select the LLM Monitoring Application tab.
Integration Types¶
On the Create Application page, integration with Langfuse is supported by default:
Access Configuration¶
Three options are supported here:
Access via Python¶
1. Install Dependencies¶
2. Initialize the Client¶
3. Verify the Connection¶
You can use langfuse.auth_check() to verify whether the integration is successful. It is not recommended for production use, as it adds latency to your application.
from langfuse import Langfuse
# Initialize (using constructor arguments)
langfuse = Langfuse(
public_key="app_id",
secret_key="llm_d9f06c320a9***6e9807",
host="https://llm-openway.guance.com"
)
# Verify the connection. This is a synchronous call and is not recommended for production. This method is only for validating the configuration.
if langfuse.auth_check():
print("Langfuse client is authenticated and ready!")
If the integration fails, you may encounter the following error:
langfuse.api.resources.commons.errors.unauthorized_error.UnauthorizedError: status_code: 401, body: {}
This error usually indicates that the App ID or Client Token is incorrect. Please check your parameter configuration.
For more detailed information and end-to-end examples, see Python.
Access via JS/TS¶
Note
This configuration only applies to the Langfuse JS/TS SDK v4. Version 3 is not supported.
1. Install Dependencies¶
Click to view more package information.
2. Register Credentials¶
Set the following environment variables in your application environment. Make sure you have a .env file in your project root directory and a package like dotenv to load the variables.
LANGFUSE_SECRET_KEY="llm_122f6b85d0c7454fa0f9*****d360"
LANGFUSE_PUBLIC_KEY="pk-lf-..."
LANGFUSE_BASE_URL="https://llm-openway.guance.com"
3. Initialize OpenTelemetry¶
The Langfuse SDK tracing is built on top of OpenTelemetry, so you need to set up the OpenTelemetry SDK. LangfuseSpanProcessor is the key component for sending traces to Langfuse.
import { NodeSDK } from "@opentelemetry/sdk-node";
import { LangfuseSpanProcessor } from "@langfuse/otel";
const sdk = new NodeSDK({
spanProcessors: [new LangfuseSpanProcessor()],
});
sdk.start();
Click to view the Ollama integration example.
For more detailed information and end-to-end examples, see JS/TS.
Access via Other¶
If you need to integrate other models/frameworks, such as DeepSeek, Ollama, LangChain, Dify, etc., see the official Langfuse documentation for configuration.
LangChain OpenTelemetry Integration¶
LangChain applications can be integrated via OpenTelemetry. After the integration is complete, you can view call traces, input, output, and token consumption in LLM Monitoring.
The system parses the OpenTelemetry gen_ai attributes and displays:
- Input: The input content of the model call;
- Output: The output content returned by the model;
- Token: Input, output, and total token consumption;
- Related information such as model, service, and call status.
Note
What is displayed on the page depends on whether the reported data contains the corresponding gen_ai attributes. Fields that are not reported are not displayed, and this does not affect queries of other trace data.
Fill in the Variables in the Code¶
Fill in the following information on the right side of the page:
| Field | Description |
|---|---|
| Application Name | Custom application name |
| Application ID | Public key. Only letters, numbers, and underscores are supported. Used to attribute data to the current application in traces, logs, and metrics. |
| Service Address | The Dataway gateway address |
| Client Token | Automatically generated after the application is created. Used to send data from user devices. |
What is a Client Token?
After you create an application, the system automatically generates a Client Token for you. This token is used to send data from user devices.
For more details, see Client Token Management.
After confirming that the information is correct, click Create to complete the LLM Monitoring application integration.