Integrating Langfuse with Embabel Agents

This guide shows how to integrate Langfuse with Embabel using OpenTelemetry.

Embabel: JVM framework created by Rod Johnson (founder of Spring Framework) for building agentic AI applications. It features a modular planning system (GOAP, Utility AI, state machines) that allows agents to dynamically decide their actions, and includes zero-code observability via OpenTelemetry.

Langfuse: Open source LLM engineering platform for observability, evals and prompt management.

Step 1: Set Up Langfuse

• Sign up for Langfuse Cloud or self-host Langfuse.
• Create a new project in Langfuse and copy your Public Key and Secret Key.

Step 2: Add Dependencies

Add the Embabel observability starter and the Langfuse OpenTelemetry exporter to your pom.xml:

First, set the Embabel version property in your pom.xml:

<properties>
    <embabel-agent.version>0.3.4</embabel-agent.version>
</properties>

Then add the following dependencies:

<dependencies>
    <!-- Embabel Agent Observability Starter -->
    <dependency>
        <groupId>com.embabel.agent</groupId>
        <artifactId>embabel-agent-starter-observability</artifactId>
        <version>${embabel-agent.version}</version>
    </dependency>
 
    <!-- Langfuse OpenTelemetry Exporter (you can use Zipkin too) -->
    <dependency>
        <groupId>com.quantpulsar</groupId>
        <artifactId>opentelemetry-exporter-langfuse</artifactId>
        <version>0.4.0</version>
    </dependency>
</dependencies>

Step 3: Configure Observability

Add the following to your application.properties:

# Enable observability
embabel.observability.enabled=true
embabel.observability.service-name=observability
 
# Enable Spring Boot tracing
management.tracing.enabled=true
management.tracing.sampling.probability=1.0
 
# Langfuse
management.langfuse.enabled=true
management.langfuse.endpoint=https://cloud.langfuse.com/api/public/otel
management.langfuse.public-key=pk-lf-…
management.langfuse.secret-key=sk-lf-…

Replace pk-lf-… and sk-lf-… with your actual Langfuse project keys. If you self-host Langfuse, update the endpoint URL accordingly.

Step 4: Run a Test

Start your application:

mvn spring-boot:run

Trigger an agent execution — for example, by calling an endpoint or running a CommandLineRunner that invokes an Embabel agent. You should see trace data being exported in the application logs.

Step 5: Explore Traces in Langfuse

Once traces are flowing, open the Langfuse dashboard:

• Trace list: See all agent executions with timestamps, latency, and token counts.
• Trace timeline: Drill into a single trace to see the full span hierarchy — agent actions, LLM calls, tool invocations, and planning steps.
• Input/Output details: Inspect prompts sent to the model and completions received, along with metadata like model name and token usage.
• Analytics dashboards: Track cost, latency, and usage trends over time.

Troubleshooting

No Traces:

• Verify that embabel.observability.enabled is set to true.
• Check that the OTLP endpoint and credentials are correct.
• Confirm that an agent action has actually been executed.

Missing Prompt/Completion:

• Ensure the sampling probability is set to 1.0 during development.
• Check that the Langfuse OpenTelemetry exporter dependency is on the classpath.

GitHub Discussions