Skip to main content

​
The Problem

AI models have a limited context window. As conversations grow long, earlier messages may be pushed out, causing the AI to lose track of important context from earlier in the discussion.

​
How Compaction Works

Raycaster Doc uses chat compaction to summarize long conversations while preserving the key information the AI needs to continue working effectively. When a conversation approaches the context limit:
1

Compaction triggers

The system detects that the conversation is approaching the model’s context window limit.
2

Memory extraction

A special turn runs where the AI is asked to write down everything it needs to remember — key facts, decisions, file references, and current task state.
3

Compaction message saved

The summary is stored as a compaction message in the conversation.
4

History windowed

Future turns only include messages after the most recent compaction point, plus the compacted summary as context.

​
When Does It Happen?

Compaction can trigger in two ways:
  • Proactive — Automatically when the conversation approaches the context limit
  • Manual — You can trigger compaction from the chat menu if you want to compress the conversation

​
What Gets Preserved

The compaction summary captures:
  • Key facts and decisions from the conversation
  • File references and workspace state
  • Current task progress and next steps
  • Important user preferences mentioned during the chat

​
Impact on Your Experience

Compaction is designed to be seamless:
  • The conversation continues normally after compaction
  • You can still scroll back and read earlier messages in the UI
  • The AI’s responses should remain contextually appropriate
  • A compaction marker appears in the chat timeline