Example: Exporting OpenTelemetry (OTLP) Traces from the TensorZero Gateway

This example shows how to export traces from the TensorZero Gateway to an external OpenTelemetry-compatible observability system.

Here, we'll export traces to a local instance of Jaeger.

Getting Started

Setup

1. Install Docker.
2. Generate an API key for OpenAI (OPENAI_API_KEY).
3. Set the OPENAI_API_KEY environment variable.
4. Launch the TensorZero Gateway, Postgres, and Jaeger: docker compose up

Running the Example

First, let's make an inference request to the gateway.

curl -X POST "http://localhost:3000/openai/v1/chat/completions" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "tensorzero::model_name::openai::gpt-4o-mini",
    "messages": [
      {
        "role": "user",
        "content": "Write a haiku about TensorZero."
      }
    ]
  }'

Then, let's make a feedback request to the gateway.

[!IMPORTANT]

Make sure to replace the episode_id with the actual episode ID from the inference request above (not the inference ID!).

curl -X POST "http://localhost:3000/feedback" \
  -H "Content-Type: application/json" \
  -d '{
    "metric_name": "comment",
    "episode_id": "00000000-0000-0000-0000-000000000000",
    "value": "Great haiku!"
  }'

Finally, visit the Jaeger UI at http://localhost:16686 to see the traces.