Skip to main content

​
Overview

This guide shows you how to integrate Latitude Telemetry into a program built with DSPy. DSPy has no dedicated instrumentor — it routes every language-model call through LiteLLM. Instrumenting LiteLLM therefore captures all of DSPy’s model calls, including those issued by modules like Predict and ReAct.
You’ll keep writing DSPy modules exactly as you do today. Telemetry observes the model calls DSPy makes under the hood.
DSPy instrumentation is available in the Python SDK only, via the LiteLLM integration.

​
Requirements

  • A Latitude account and API key
  • A Latitude project slug
  • A project that uses DSPy (dspy, which depends on litellm)
  • A key for whichever provider your dspy.LM targets (e.g. OPENAI_API_KEY)

​
Steps

1

Install

pip install latitude-telemetry dspy litellm
2

Initialize and use

import dspy
import litellm

from latitude_telemetry import Latitude, capture

# DSPy routes every LM call through litellm, so instrumenting litellm
# captures DSPy's model calls.
latitude = Latitude(
    api_key="your-api-key",
    project="your-project-slug",
    instrumentations={"litellm": litellm},
)

dspy.configure(lm=dspy.LM("openai/gpt-4o-mini"))

def dspy_qa():
    qa = dspy.Predict("question -> answer")
    return qa(question="Hello").answer

capture("dspy-qa", dspy_qa)

latitude.shutdown()

​
Seeing Your Traces

Once connected, traces appear automatically in Latitude:
  1. Open your project in the Latitude dashboard
  2. Each execution shows input/output messages, model, token usage, latency, and errors
  3. Wrap a request with capture() to group every model call DSPy makes into one trace