Keyboard shortcuts

Press or to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Chapter 22 Project Practice: Multimodal Agent

🎨 "The real world speaks more than words — images, audio, and video are all languages an Agent needs to understand."


🎓 Learning Objectives

After completing this chapter, you will be able to:

  • ✅ Understand the capability boundaries and application scenarios of multimodal Agents
  • ✅ Master integration patterns for vision understanding and image analysis
  • ✅ Integrate speech input/output with an Agent
  • ✅ Build a multimodal personal assistant with image and voice capabilities
  • ✅ Understand the core architecture and safety practices of Computer Use / GUI Agents
  • ✅ Master implementation methods for video understanding and multimodal RAG

⏱️ Estimated Learning Time

Approximately 150–180 minutes (capstone project)

💡 Prerequisites

  • Complete all chapters from Part I to Part V
  • Understand base64 encoding and multipart HTTP requests

Chapter Overview

In previous chapters, all the Agents we built only processed text. But in the real world, information exists in many forms — users might send a screenshot asking "how do I fix this error," or say "help me analyze this chart." Multimodal Agents can understand and generate multiple types of content, greatly expanding application scenarios. This chapter will guide you to build a multimodal personal assistant from scratch that can handle text, images, and audio simultaneously.

Chapter Goals

After completing this chapter, you will be able to:

  • ✅ Understand the core capabilities and application scenarios of multimodal models
  • ✅ Implement image understanding and analysis with GPT-4.1
  • ✅ Implement image generation with DALL-E
  • ✅ Integrate speech recognition (STT) and text-to-speech (TTS)
  • ✅ Build a complete multimodal personal assistant Agent
  • ✅ Understand Computer Use / GUI Agent architecture and safety practices
  • ✅ Implement video understanding and multimodal RAG

Chapter Structure

SectionContentDifficulty
22.1 Multimodal Capabilities OverviewCapabilities and application scenarios of multimodal models⭐⭐
22.2 Image Understanding and GenerationGPT-4.1 analyzes images, DALL-E generates images⭐⭐⭐
22.3 Voice Interaction IntegrationSpeech recognition and text-to-speech⭐⭐⭐
22.4 Practice: Multimodal Personal AssistantBuild a complete multimodal Agent⭐⭐⭐⭐
22.5 Computer Use and GUI AgentsLet Agents operate computers and browsers⭐⭐⭐⭐
22.6 Video Understanding and Multimodal RAGVideo analysis and hybrid image-text retrieval⭐⭐⭐⭐

⏱️ Estimated Learning Time

Approximately 90–120 minutes (including hands-on exercises)

💡 Prerequisites

  • Completed Agent development basics from previous chapters
  • Familiar with basic OpenAI API usage (including Vision and Audio APIs)
  • Python async programming basics (async/await)

🔗 Learning Path

Core Prerequisites: Chapter 3: Tool Use / Function Calling, Chapter 12: LangGraph Recommended but not required: Chapter 17: Agent Evaluation and Optimization

Related Projects:


Next: 22.1 Multimodal Capabilities Overview