LlamaIndex Function Agent integrated with ZenML
LlamaIndex lets you build function agents that call multiple tools and often run asynchronously; integrating it with ZenML executes those agents inside reproducible pipelines with artifact lineage, observability, and a clean path from local development to production.
Features with ZenML
- Async-friendly orchestration. Run LlamaIndex function agents that require awaiting inside ZenML steps without changing agent code.
- Tool call lineage. Track queries, intermediate tool outputs, and final responses as versioned artifacts.
- Composable pipelines. Chain agents with retrieval, evals, and deployment in one DAG.
- Evaluation ready. Add post-run checks to score response quality, latency, and tool accuracy.
- Portable execution. Move the same pipeline from local runs to Kubernetes or Airflow via ZenML stacks.
Main Features
- Function agents. Define agents that call Python tools to solve tasks.
- Async execution. Properly await
agent.run(...) for non-blocking workflows. - Multiple tools. Plug in weather, tip calculator, and custom utilities.
How to use ZenML with
LlamaIndex
from zenml import ExternalArtifact, pipeline, step
from agent import agent # LlamaIndex function agent with tools
@step
def run_llamaindex(query: str) -> str:
# LlamaIndex agent.run is async; await it inside the step
import asyncio
async def _run():
return await agent.run(query)
resp = asyncio.run(_run())
return str(getattr(resp, "response", resp))
@pipeline
def llamaindex_agent_pipeline() -> str:
q = ExternalArtifact(
value="What's the weather in New York and calculate a 15% tip for $50?"
)
return run_llamaindex(q.value)
if __name__ == "__main__":
print(llamaindex_agent_pipeline())
Additional Resources
ZenML Agent Framework Integrations (GitHub)
ZenML Documentation
LlamaIndex Documentation
