Skip to main content
🎤 Luca Berton is speaking at Red Hat Summit & KubeCon EU 2026!Learn more →
intermediate

Build Durable AI Agents with Dapr Agents

Agents that survive restarts: workflows, state and pub/sub with Dapr Agents, from the first agent to a multi-agent system.

6 lessons35mUpdated November 2026

What you'll learn

  • Install Dapr in slim mode, without containers, and make a first model call through a conversation component
  • Build a durable agent that finishes its task after its process is stopped halfway
  • Write workflows that mix plain steps with a model step, with a retry policy
  • Serve an agent over HTTP, then start, poll, terminate and purge its instances from a client
  • Connect an order agent, a policy agent and an orchestrator through pub/sub with Redis
  • Read the traces of the agent and its sidecar, keep the model key in a secret store, and add retries for the model

Who this is for

  • • DevOps engineers deepening their expertise
  • • SREs building production-grade systems
  • • Backend developers expanding into infrastructure

Prerequisites

  • • Comfortable with Python and the command line
  • • A Mac or a Linux machine with Python 3.11 to 3.14, curl, jq and Redis
  • • An OpenAI-compatible model endpoint and its key

Course content

2 modules, 6 lessons, 35m of video

Module 1

3 lessons, 16 min
1

6 min

The learner installs Dapr and its slim runtime to run services without containers. They then call a model using a configuration component, demonstrating that the model is defined in a YAML file.

2
A durable agent that survives a restart

6 min

The learner writes a tool and a durable agent served over HTTP, then stops and restarts the process to demonstrate durability. They learn that the workflow history survives the process, and interrupted steps run again without repeating finished steps.

3
Workflows that mix model steps and plain steps

5 min

The learner builds a workflow that combines plain activities and a model activity, applying a retry policy to the model step. They learn that the workflow function is replayed from its history, and that model calls are limited to where judgment is needed.

Module 2

3 lessons, 19 min
4
The agent as an HTTP service

5 min

This lesson explains that a durable agent functions as an ordinary HTTP service, allowing clients to start, poll, terminate, and purge task instances using standard HTTP calls. Learners read the agent's OpenAPI description and see how tasks are managed as individual workflow instances with associated IDs and statuses.

5
Two agents talking through pub/sub

6 min

The learner runs a small team of agents that communicate using pub/sub messaging, without direct connections between them. They learn that a topic acts as an agent's address, and the registry allows an orchestrator to find the team members.

6
See what the agent did, then harden it

9 min

The learner traces an agent's activity, secures its model key using a secret store, and configures retries for a failing model component. This demonstrates how to record agent actions and harden the system against failures.

Get started

Start Free Lesson

First lesson is always free — no signup required

This course includes

6 hands-on lessons
Intermediate level
35m total
Final assessment

Want all courses?

See Pro plan →

Your instructor

Luca Berton

Founder and CEO, Open Empower. About the author

More in AI Agents

All courses in this track