Build a Research Assistant AI Agent with TypeScript (Part 2): Callbacks, State, and Memory

This is Part 2 of a 2-part series. If you haven’t built the core pipeline yet, start with Part 1. It covers the Sequential Agent pattern, shared state, and the 4-agent research pipeline we’re building on here. You can find the full source code for this project on GitHub. In Part 1, we built a research assistant using ADK-TS’s Sequential Agent pattern: four agents running in a strict pipeline, communicating through shared state. The result was a working system that takes any topic and produces a comprehensive research report. But “working” and “production-ready” are two different things. Right now, our pipeline runs silently. It kicks off while we wait around 30-60 seconds, with no idea what’s happening. There’s no way to configure the behavior without editing the code. And every research session vanishes the moment it completes. Shipping it like this might feel faster, but it’s the kind of shortcut that quickly becomes technical debt. In this guide, we’ll add four features from ADK-TS,