The HoneyHiveLangChainTracer class is a utility designed to trace and log LangChain operations with the HoneyHive API. It extends the BaseCallbackHandler from LangChain and provides methods for session management and updating various properties.

​
Explanation

The HoneyHiveLangChainTracer is designed to capture and log various events that occur during the execution of LangChain operations. It does this by overriding several “handle” methods from the BaseCallbackHandler class. These methods are automatically called by LangChain at different points during the execution of chains, LLMs, tools, and other components.

You can find the specific implementation of those handler functions in this file.

​
Constructor

​
constructor(input: HoneyHiveTracerInput)

Initializes a new instance of the HoneyHiveLangChainTracer.

Parameters:

  • input: HoneyHiveTracerInput - An object with the following properties:
    • project: string - Name of the project associated with this tracing session.
    • sessionName: string - Name for this specific session.
    • source?: string - Source identifier (default: ‘langchain’).
    • userProperties?: Record<string, any> - User properties for the session.
    • metrics?: Record<string, any> - Initial metrics for the session.
    • config?: Record<string, any> - Configuration for the session.
    • metadata?: Record<string, any> - Initial metadata for the session.
    • apiKey?: string - API key for authenticating with HoneyHive (falls back to HONEYHIVE_API_KEY environment variable if not provided).
    • verbose?: boolean - Whether to log verbose output (default: false).
    • baseUrl?: string - HoneyHive API base URL (default: ‘https://api.honeyhive.ai’).

Throws:

  • Error if the HoneyHive API key is not set.

Usage Example:

const tracer = new HoneyHiveLangChainTracer({
  project: "MyProject",
  sessionName: "TestSession",
  apiKey: "your-api-key-here"
});

​
Methods

​
async setFeedback(feedback: Record<string, any>): Promise<void>

Sends feedback to HoneyHive, associating it with the current session.

Parameters:

  • feedback: Record<string, any> - Feedback to be sent to HoneyHive.

Returns: Promise<void>

Usage Example:

await tracer.setFeedback({ rating: 5, comment: "Excellent response" });

​
async setMetric(metrics: Record<string, any>): Promise<void>

Sends metrics to HoneyHive, associating them with the current session.

Parameters:

  • metrics: Record<string, any> - Dictionary of metrics to be sent to HoneyHive.

Returns: Promise<void>

Usage Example:

await tracer.setMetric({ response_time: 1.5, accuracy: 0.95 });

​
async setMetadata(metadata: Record<string, any>): Promise<void>

Sends metadata to HoneyHive, associating it with the current session.

Parameters:

  • metadata: Record<string, any> - Dictionary of metadata to be sent to HoneyHive.

Returns: Promise<void>

Usage Example:

await tracer.setMetadata({ user_id: "12345", context: "customer_support" });

​
async setUserProperties(userProperties: Record<string, any>): Promise<void>

Updates user properties for the current session in HoneyHive.

Parameters:

  • userProperties: Record<string, any> - Dictionary of user properties to be sent to HoneyHive.

Returns: Promise<void>

Usage Example:

await tracer.setUserProperties({ user_type: "premium", language: "en" });

​
async startNewSession(): Promise<void>

Starts a new tracing session with HoneyHive.

Returns: Promise<void>

Notes:

  • This method uses the configuration provided in the constructor to initialize a new session.
  • It sets the sessionId property of the tracer instance.

Usage Example:

await tracer.startNewSession();

​
Error Handling

All methods log errors to the console if they fail. If verbose is set to true in the constructor, additional error details will be logged.

​
Notes

  • The HoneyHiveLangChainTracer class automatically traces various LangChain operations (LLM calls, chain executions, tool usage, etc.) and sends the traces to HoneyHive.
  • Make sure to call startNewSession() before using other methods if you want to explicitly start a new session. Otherwise, the session will be implicitly started when the first trace is sent.
  • The sessionId is automatically generated and managed by the tracer. You don’t need to provide or manage it manually.
TypeScriptPython