Session state is a dictionary where your agent stores data it needs to remember and access programmatically. Unlike conversation history (which only the LLM can read), session state gives you code-level access to exact values.

The problem: Conversation history alone isn’t enough for programmatic control

The solution:  Session state

Quiz Question 1

Quiz Question 2

Quiz Question 3

Session State Fundamentals

In Part 1, you learned that session state is a dictionary you can access programmatically. Now you'll learn to inject state values into agent instructions using {var} templating, enabling dynamic agent behavior.

The problem: How do agents use state values in their instructions?

The solution: State Templating with {var}

Dynamic Instructions with State

You've learned to use state for programmatic control and data passing. Now you'll master state namespaces—controlling how long state values persist and who can access them.

The problem: Not all state should persist equally

The solution: State Namespaces

State Namespaces and Persistence

Reading list

Reading List

You've built basic LLM agents that respond to queries—now let's make them stateful. Use session state to build agents that maintain context, remember user preferences, and provide personalized experiences. Transform agents from stateless responders to intelligent assistants.

Build stateful AI agents using session state and memory management. Create agents that maintain context and remember user preferences.

Google: Manage Agent Memory and State

Build stateful AI agents that maintain context and remember user preferences using session state, memory management, and personalization.

The solution: Session state

Google: Manage Agent Memory and State

动手互动练习