Building Agentic Applications Using Angular and Genkit

Build a complete AI agent with a streaming Angular front end and a Genkit back end: tools, memory, retrieval with citations, planning, human approval, multi-agent delegation, evaluation and deployment. One working application, explained chapter by chapter, with every code listing checked against the repository.

What you will learn

• Define an agent's contract once in Zod and share it between the server and the Angular page
• Stream typed events over fetch and fold them into signal state with a pure, testable reducer
• Write Genkit tools that act as a particular user, keep rules in code and return errors the model can use
• Save conversations so they survive restarts, trim history safely, and answer questions from a document with citations
• Add planning and self-checking, and measure what they cost in model calls and tokens
• Let an agent answer with Angular components without ever letting a model write markup
• Pause a tool call for human approval, resume it exactly once, and limit what an approver can change
• Split work across specialist agents, and decide whether the team is worth its cost
• Evaluate the agent with Genkit evaluators, break it on purpose, then rate-limit, validate, log and deploy it.

Who this book is for

Web developers who have built something in Angular and now need to add an agent to a product, and back-end TypeScript developers who want to know what the front end of an agent has to do. You should be comfortable with TypeScript. No API key is required: the example application includes an offline scripted model, so every chapter runs on a laptop and every test is deterministic.