LlamaIndex Workflow: Build Practical AI Agents Fast

Transcript

1. 0 LlamaIndex Workflow: Build Practical AI Agents Fast 2025-10-24 第133回NearMe技術勉強会 Cyan Chen
2. 1 Why LlamaIndex, Why Now • Fragmented LLM stack → need coherent data + agent layer • LlamaIndex = opinionated toolkit for RAG and tool-using agents and workflows • Strengths: rich data connectors, retrieval engines, agent runtimes, callbacks • Outcome: faster prototypes → production with less glue code • Think of why we need Expressjs, Nestjs, Nextjs for web development
3. 2 LlamaIndex Workflow at a Glance • A Workflow in LlamaIndex is an event-driven abstraction used to chain together several events. • Workflows are made up of steps. • Steps handle events and emit events. • Events in Python SDK are Pydantic models. • Each workflow must define exactly one StartEvent. • The entrypoint could be StartEvent or an event that extends StartEvent. • A workflow stops when a StopEvent is emitted. https://developers.llamaindex.ai/python/workflows/
4. 3 Drawing a Workflow Git: https://github.com/yukimotochern/llam a_index_study
5. 4 Workflow Server • They didn’t release this feature 2 months ago while I was evaluating LlamaIndex for NearMe, so I picked a custom way to expose each workflow as a MCP server tool. • Now, let’s try this.
6. 5 Waiting for Multiple Events
7. 6 Manually Triggering Events
8. 7 Streaming Events
9. 8 AG-UI: the Agent-User Interaction Protocol https://github.com/ag-ui-protocol/ag-ui
10. 9 Analyze AGUIChatWorkflow
11. 10 Thank you