Skip to main content

Chat Compaction

As conversations grow longer, Revelry Ai uses chat compaction to summarize older messages. This keeps conversations manageable while preserving important context.

What is Compaction?​

Compaction automatically summarizes older messages in a conversation to manage the AI model's context window. Instead of sending the entire conversation history with every message, compacted summaries replace older messages with a concise recap.

This allows you to have long, productive conversations without losing important context or hitting model limits.

When Does Compaction Happen?​

Compaction triggers automatically based on the AI model's context thresholds. You don't need to configure anything — the system monitors conversation length and compacts when needed.

Different AI models have different context window sizes, so the compaction threshold varies by model.

What You'll See​

When compaction occurs, you'll see a compaction status card inline in the conversation:

  • Pending — Compaction is in progress, the system is generating a summary
  • Completed — The summary has been created and older messages have been compacted

The status card appears where the compacted messages were in the conversation timeline.

What's Preserved in Summaries​

Compaction summaries are designed to retain the most important context:

  • Key decisions made during the conversation
  • Artifact trail — references to any artifacts created or discussed
  • Pending items — open questions or tasks that haven't been resolved
  • Session intent — the overall goal or direction of the conversation

Cancelling a Compaction​

If you see a compaction in Pending status, you can cancel it before it completes. This keeps the original messages intact without summarization.

Once a compaction reaches Completed status, it cannot be undone.

Compaction in Workflows​

Chat compaction also works within workflow steps. When a workflow step involves a long-running chat execution, compaction automatically manages the context window so the step can complete without hitting model limits. This is especially useful for complex workflows that involve extensive tool calling or large amounts of data.

Tips​

  • Long conversations are fine — Compaction handles context management automatically
  • Important details are preserved — The AI prioritizes keeping key decisions and context
  • Continue naturally — After compaction, keep chatting as normal. The AI has the summarized context available

Next Steps​