Customer-facing AI agents respond generically by default. They don't know what the user has been doing on your site — what pages they've viewed, how long they've been browsing, what they've been comparing. Every response starts from zero.

This accelerator shows you how to close that gap. It walks you through adding Snowplow Signals to a Google ADK agent so it has full session context before every response. Google's Agent Development Kit is an open-source framework built to make agent development feel like software development — modular, composable, and deployable to Vertex AI Agent Engine. CopilotKit handles the UI wiring, connecting the agent to a Next.js front-end. Signals feeds live behavioral attributes into the system prompt on every turn.

The result is an agent that knows, in real time, what the user has been doing before it says a word.

‍

Use Cases

• In-session product assistant — personalize agent responses based on what a user is actively browsing, comparing, or spending time on in the current session
• Contextual support agent — give support agents full behavioral context before the first message, so users don't have to explain what they were doing when the issue occurred
• Editorial recommendation agent — adapt content suggestions in real time based on which articles, categories, or topics a user has engaged with‍
• Onboarding assistant — surface relevant next steps based on which parts of a product or docs a new user has already explore

‍

Infrastructure Overview

• Snowplow JavaScript Tracker — captures behavioral events (page views, article interactions, session data) from the Next.js front-end application
• Snowplow Signals — processes raw event data into behavioral attributes and serves them via the Profiles API; fetched at runtime to understand what the user is doing right now
• Google ADK (Gemini) — open-source agent framework that handles orchestration, tool use, and model interaction; optimized for Gemini and deployable to Vertex AI Agent Engine
• — handles UI wiring between the ADK agent and the Next.js front-end, injecting fresh session context into the system prompt each turn

‍

Additional Insights

The stack is composable by design — swap any layer without touching the others. Teams already running ADK can add Signals for real-time context without changing their agent orchestration logic. The pattern of injecting live behavioral attributes into the system prompt on every turn is framework-agnostic; the same Signals Profiles API call works with LangChain, LlamaIndex, or any other ADK-compatible framework.