Conversation
The full running thread of turns the model reads to keep a chat coherent.
Think of it like
The whole email chain, not just the latest reply — context you’d lose if you only read the bottom message.
Example
When a chatbot recalls something from ten messages ago, it’s because the whole conversation is re-fed each turn.
How it actually works
A conversation is the ordered stack of system, user, and assistant turns. Models are largely stateless, so the app resends the accumulated history every call to simulate memory. That history grows until it bumps the context limit, forcing truncation or summarization.
For product teams
The model’s “memory” is really just resent history — real long-term memory needs extra machinery.
For engineers
The concatenated sequence of turns re-sent per request; statelessness simulated by replaying history.
Related
- Turn — Made of turns.
- Context length — Bounded by context length.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome