In words
What it is, why it matters, and what it is like.
Why am I learning this?
Every time you chat with an AI assistant, it appears to remember your previous messages. That memory is not magic; it is a specific technical constraint that shapes how the system works. In this page you will learn exactly what information is stored and sent forward each time you type a message, why the system cannot keep every word of every conversation forever, and how summarizing, extracting key facts, and tracking tasks helps manage that limit. Understanding this process is essential for using AI agents effectively — systems that plan, call external tools, and perform multi-step work. If you do not understand how memory works here, your agents will forget who you are, repeat themselves, or fail at complex tasks because they lost the thread of what was discussed.
The idea, in plain terms
Imagine you are a customer service agent at a company. A customer calls, and you have only a small notepad. As the conversation goes on, you write down the key points: 'Customer's name is Priya', 'Problem with order #12345', 'She wants a refund'. You cannot write the whole conversation — there is no space and you would drown in detail. So you write a summary, you note the important facts, and you keep a list of what needs to happen next. When the next person takes over the call, you hand them the notepad. That notepad is the conversation state. In AI systems, this 'notepad' is a collection of data that stores the history of messages, a compressed summary, extracted facts, and a list of pending actions. Every time the model needs to respond, it reads this notepad and sends it back along with your new message. This notepad has a size limit — the maximum amount of text it can process at once — so you must decide what to keep and what to throw away. If you throw away something important, the model can no longer reason about it because that information is gone.
An analogy
Think of the AI's memory like a whiteboard in a classroom. You and the AI are solving a problem together on this board. The AI reads whatever is written there to understand where things stand. But the whiteboard is only so big — you cannot write an entire novel on it. So, as new information comes in, old details fade or get wiped off to make room for what matters now. The AI might keep a list of key facts ('Priya's allergy: nuts') and a to-do list ('Send recipe'), but the full transcript of what was said five minutes ago is gone unless it was specifically saved to a separate note. The rule of this analogy is that if information is not currently on the whiteboard (the active state), the AI has no way to know it exists. It does not have an intuition or a gut feeling; it only knows what is written in front of it. That is why managing what goes on the board is critical.
Definition
Conversation state is the specific set of information — including message history, summaries, facts, and tasks — that represents the current context of a dialogue and is provided to the AI model each time it generates a response. This set of information is constrained by the context window, which is the maximum amount of text the model can process in one turn.
Where this sits
This concept is the foundation for understanding AI agents, which are systems that perform multi-step actions rather than just answering questions. Conversation state provides the 'memory' that allows an agent to connect today's action to yesterday's plan. It also directly relates to the context window, which is the physical limit on how much information can be in that state at any given time. While conversation state defines *what* is remembered, the context window defines *how much* of that memory can fit into the model's immediate processing space.
