Make DeepAgents Durable with Dapr Workflow - Deep Issue Investigation
Duration: 30 minutes
Supported languages: Python. Requires an OpenAI API key.
A real issue investigation isn't one LLM call — it's dozens of tool calls and reasoning steps, and it can take minutes. All of that lives in memory by default: kill the process at step 30 of 40 and you lose the scratchpad, the partial report, and every dollar spent on LLM calls so far. In this self-paced track you'll see how Dapr Workflow turns a DeepAgents agent into a durable, fault-tolerant application.
What you'll run
You'll work with a Python CLI tool that uses a DeepAgents agent to investigate a real Dapr bug — dapr/dapr#7326 — using a local GitHub snapshot, and write a Markdown investigation report. You'll run the in-memory baseline first, then make it durable with the DaprWorkflowDeepAgentRunner so every tool call is checkpointed to a Dapr state store. Finally you'll crash the investigation mid-run and watch the workflow resume from the last checkpoint — replaying the completed steps instead of calling the LLM again. You'll need around 30 minutes to complete the 4 challenges.
Ensure you have your own OpenAI API key before you start this track.
In this self-paced track, you'll learn:
- Why durable execution matters when an agent performs a long, multi-step investigation with dozens of LLM and tool calls.
- How to run a DeepAgents agent that investigates a real GitHub issue and writes an investigation report.
- How to wrap the agent in a Dapr Workflow so every tool call becomes a checkpointed activity backed by a Dapr state store.
When you've registered to access learning content, click the View fullscreen button in the course player below.
Having trouble loading the course? Open in a new tab