Context Compression
Domain: Memory & Context
Motivation
When a conversation gets very long, the agent compresses older messages into a summary to fit within its context window. You keep the meaning but use less space.
Concrete Example
After 50 messages, the agent compresses messages 1-40 into: User asked about Python lists, dicts, and loops. Decided to build a calculator. Now the context has room for new messages.
Analogy
Summarising a long meeting into bullet points - instead of keeping the full 2-hour transcript, you write 5 key points. Same information, much less space.