Memory & conversation state
Build the mental model
Chat models do not remember yesterday by themselves. Memory is an application system for writing, storing, retrieving, and forgetting state across turns and sessions.
1Try it yourself
Playground
Conversation memory vs context window
Chat history eats tokens. Summarize or trim old turns to stay in budget.
Context used: 3 / 5 slots
2Learn the idea
Read
Analogy for this concept only
Think of a notebook the application manages: decide what deserves writing, store it with provenance, and decide what to read back later. Use the analogy to name the moving parts for conversation memory, then drop it when you need numbers. For the personal tutoring assistant, the enduring idea is not a vendor feature name; it is the decision conversation memory changes and the evidence that decision leaves behind.
Chat models do not remember yesterday by themselves. Memory is an application system for writing, storing, retrieving, and forgetting state across turns and sessions.
Beginners often blur neighboring ideas when discussing conversation memory. Keep it distinct by asking what artifact would still exist if model weights were frozen and only this layer changed on the personal tutoring assistant. If you cannot name that artifact, you are still describing “the AI” in general.
Read
Case lens: personal tutoring assistant
Extract candidate memories, validate/consent, store with metadata, retrieve relevant items into the prompt, update or expire them, and keep audit trails. In day-to-day language for conversation memory: someone brings a need, the system inspects allowed evidence, this layer contributes a judgment or structure, and a consequence reaches a user or downstream system. Deterministic guards—permissions, schemas, arithmetic—still belong to the application around the personal tutoring assistant.
Uncertainty is normal for conversation memory. Incomplete inputs and probabilistic behavior mean the personal tutoring assistant needs an escape hatch (retry, fallback, escalate) rather than fake certainty in fluent prose.
Read
Make it operational
When you explain conversation memory to a new teammate on the personal tutoring assistant, forbid the sentence “the AI just knows.” Replace it with the artifact that moves and the evidence you would file for conversation memory. If they can falsify your picture with a single counterexample from last week’s traffic on the personal tutoring assistant, your mental model is working.
Also pin one numeric memory from this conversation memory chapter: If 40 memories are injected every turn but only 3 affect the hint, you spent tokens and risked contradictions—measure useful-memory rate, not memory count. That number is not decoration; it is a template for how claims about conversation memory on the personal tutoring assistant should look in design docs. Scoped specifically to conversation memory / personal tutoring assistant / mental-model.
Read
Common mix-ups
People confuse conversation memory with neighboring buzzwords when debugging the personal tutoring assistant. Before changing prompts, ask whether the broken stage was evidence gathering, the conversation memory judgment itself, validation, or the product action. Fixing the wrong stage creates folklore (“we tried conversation memory and it failed”) that blocks the next team on the personal tutoring assistant. Scoped specifically to conversation memory / personal tutoring assistant / mental-model.
Go deeper
Before you start
Why this matters
Spend two minutes on the personal tutoring assistant. If conversation memory disappeared tomorrow, what breaks first for the user, and what evidence would prove it was working? Write that before you read the analogy.
Related lessons
Check your understanding
Page assessment
Answer from memory. Completion is saved from this evidence, not from opening the next page.
All responses are required.