Agents Observability Demo

With LlamaIndex and OpenTelemetry!

This demo showcases a solution for agent observability and tracing by building on the LlamaIndex x OpenTelemetry integration.

It traces the activity of an agent capable of extracting information from a University course syllabus (with LlamaExtract) and answer questions about it (with a LlamaCloud Index in the backend).

The agents tool are served via MCP!

Get it up and running!

Get the GitHub repository:

git clone https://github.com/run-llama/agents-observability-demo

Install dependencies:

cd agents-observability-demo/
uv sync

And then modify the .env.example file with your API keys and move it to .env.

mv .env.example .env

Now, you will have to execute the following scripts:

uv run tools/create_llama_extract_agent.py
uv run tools/create_llama_cloud_index.py

You're ready to set up the app!

Launch a local Postgres database + a local Jaeger instance to record OpenTelemetry traces:

docker compose up -d

Run the MCP server:

uv run src/agents_observability_demo/server.py

In a separate window, run the websocket:

uv run src/agents_observability_demo/websocket.py

Last, run the Gradio frontend, and start exploring at http://localhost:7860:

uv run src/agents_observability_demo/main.py

Contributing

Contribute to this project following the guidelines.

License

This project is provided under an MIT License.