Why Do I Need Traces?

Traces capture the whole journey from the user's initial query through all the processing stages to the final result. A trace typically includes timestamps, the query and how its transformed at every stage, embeddings generated, documents retrieved, scores, ranking decisions, and errors or latency issues encountered.

AI-powered applications tend to be complex and opaque; when something goes wrong it's difficult to understand why. Traces help you see exactly what happens behind the scenes, which is useful for debugging issues, optimizing performance, and understanding how your AI models are behaving. They help you identify bottlenecks causing high latency, debug queries that return unexpected results, and gather insights about search patterns.

Tracing Your deepset Pipelines

You can use the following resources to trace your pipelines:

• Pipeline logs: Capture operational details, such as the component that generated the log, log level, timestamp, and log message.
• Integrated services:
  • Langfuse: Provides deep tracing for application-level observability, including detailed spans, latency tracking, and cross-service dependencies.
  • Weights & Biases Weave: Captures ML-specific telemetry to help you track your pipeline's latency, token-count, and more.

Tracing With Logs

Logs are enabled for every deepset pipeline by default. To view the logs:

1. In deepset AI Platform, choose Pipelines.

2. Click the name of your pipeline. This opens the Pipeline Details page.

3. Open the Logs tab.

   

You can filter the logs by the message and pipeline type or date added. You can also search for keywords in log messages.

To download logs, click Download CSV.

Tracing With Weights & Biases

For instructions on tracing with Weights & Biases, see Use Weights & Biases Services.

Tracing with Langfuse

To trace with Langfuse:

1. Add a secret to connect deepset to Langfuse.
2. Add LangfuseConnector to your pipelines.

For an overview of how it works, see Using Hosted Models and External Services.