Knowledge base

What the "Context compacted" marker means

Why long conversations get summarized, what the rose seam means, and how to read or edit the summary so answers stay on track.

What just happened

You scrolled back through a long chat and hit a dashed rose line with a small chip that reads Context compacted. Nothing is broken, nothing is gone, and nothing needs your attention. Here is what it means.

Why conversations get compacted

A model answers by reading the conversation so far, and it can take in only so much text at once. That limit is its context window.

Lexika does not wait for that limit. After each answer, it measures the text of the conversation the model would read next time. Once that passes a fixed ceiling, which is lower on models with a small context window, Lexika distills the oldest part of the conversation into a summary and hands the model that summary in place of those messages. The conversation keeps going, and the model keeps answering with an understanding of what came before, in condensed form for the oldest stretch.

Lexika does this on its own, and only in regular chats. It never compacts a Council or Agentic Research conversation, and it checks only after an AI answer, so comments alone never start it. There is no button that starts it, no setting that changes when it happens, and no switch that turns it off. What you can change is the summary itself, as the section on editing below explains.

What is preserved, and what the model reads

The summary panel, which the seam chip opens, explains the arrangement in one sentence under its title. It names the cut-off, for example March 14 at 09:20 AM, says the messages up to that moment went into this summary, and adds that the model reads the summary in place of the original messages, which stay visible in the transcript. The date is in your local time.

Two separate things are worth keeping apart.

You see The model reads
Above the seam (older messages) Every original message, in full The summary
Below the seam (newer messages) Every original message, in full Every question and answer, verbatim

Your transcript is untouched. Scroll up as far as you like, and every question you asked and every answer you got is still there, word for word. Compaction changes what Lexika hands to the model, not what the conversation shows you.

Files attached above the seam stop reaching the model. Lexika wrote the summary from the text of those messages and the files' names, not from what was inside the files, so attach a file again if you still need it.

The cut-off point is a moment in time, the timestamp of the last message the summary covers. Every question and answer sent at or before that moment sits above the seam. Every question and answer after it reaches the model as written.

Lexika can compact a conversation more than once as it grows, and the boundary moves forward each time. It writes each new summary from the previous one plus the next stretch of messages, so each summary reaches from the start of the conversation up to the new cut-off.

The signals in the interface

All of them share the same rose accent. The seam chip, the floating button, the compacting status and the summary panel also carry the same history glyph, a clock with a backwards arrow.

  • The seam row. A dashed rose line running in from both edges to the small centred chip that reads Context compacted. It sits directly after the last message the summary covers. Click the chip to read the summary.
  • The dashed edge line. Each message the summary covers carries a dashed rose rule along the transcript's right edge, as tall as the message itself. The transcript keeps a left-to-right layout in every language, so the line stays on the right in Arabic too. Together, the rules read as one continuous line, showing how far back the summary reaches. Hover it, or press and hold on a touch screen, and a small tooltip reads Context compacted. It opens nothing. On a phone, the line falls just outside the visible transcript, but the message index drawer still shows it.
  • The floating indicator. When your whole view is above the seam, a small round button with a rose outline and the history glyph appears at the bottom of the transcript, beside the jump-to-latest arrow. Hover it and a tooltip reads You are reading summarized context. Clicking it opens the summary. The first time it appears on a device, a tip titled Reading summarized context points to it, and the tip's Learn more button opens this article.
  • The compacting status. While Lexika writes a new summary, a rose pill reading Compacting context… appears at the bottom of the transcript for everyone with the chat open. When Lexika saves the summary, the pill goes away and the seam moves to the new cut-off. If the attempt fails, the pill goes away and Lexika tries again after a later answer.
  • Rose in the message index. In the Message index beside the transcript, questions inside the summarized region turn rose, so a glance tells you how far back the summary reaches. The List style, and the drawer a phone uses, also draw the dashed rose line beside each covered question.

Worth knowing

You will often see no compaction marker at all, even in a compacted conversation. A chat opens on its most recent messages, and if the summary covers none of them, the transcript has nothing to mark. The seam, the edge line and the floating indicator only appear once you have scrolled back far enough to reach a covered message. The message index can turn rose sooner, as soon as a covered question is among the messages loaded so far.

Reading the summary

The summary opens from the seam chip or the floating indicator, not from the dashed edge line.

It opens in a full-height panel that slides in from the side. The panel's title is Compacted context. Under the title comes the sentence described above with the cut-off date filled in, then the summary itself in its own scrolling area. Along the bottom sit Edit summary, covered in the next section, then What does this mean?, which opens this article, and Close.

That text is the entire thing the model reads in place of your older messages, shown with its formatting. Lexika hands it to the model as reference material, not as instructions. A summary Lexika writes is a list of short points grouped by kind, such as facts, decisions and open questions, kept to about 800 words, so smaller details can drop out.

Editing the summary

If the summary got something wrong or left out a detail you need, you can rewrite it. Anyone in the workspace the chat belongs to can do this, and the change applies for everyone in the chat.

Edit summary swaps the formatted view for a text box holding the summary's Markdown source, with a counter that shows your word count against the limit of 800. Save Changes stays unavailable while the text is empty, unchanged or over 800 words. Cancel drops your changes. A successful save shows Compacted context updated.

  • The model reads your version from the next message on. An answer already in progress keeps the old one.
  • The cut-off stays put. The same messages stay above the seam, and the original messages in the transcript stay as they were.
  • Very dense text can still be refused. Lexika also caps the summary's size in a way the word counter cannot show. Code, or a language written without spaces between words, can reach that cap before 800 words, and the save then fails.
  • Reopen the panel if the summary changes under you. If the Compacting context… status appears, or a teammate edits the same summary while you are typing, copy your text, close the panel and reopen it to edit the newer version. Saving over it replaces what the newer summary added.
  • Your edits carry forward. When Lexika compacts the conversation again, it writes the new summary from your version plus the newer messages and keeps what you wrote unless newer messages replace it. It regroups the points in its usual shape, so your formatting does not survive and minor points can drop out.
  • Editing or regenerating an old question drops the summary. If you use Edit message on a question at or before the cut-off, or Regenerate on its answer, Lexika discards the whole summary, your edits included, because it describes answers that no longer exist. The seam disappears, and the next summary starts from scratch. A chat made with Branch in new chat also starts without a summary.

Getting good results in a long conversation

  • Read the summary if an answer feels off. If the model seems to have lost a detail from early on, open the summary. If the detail is not in there, restate it in your next message, or add it with Edit summary so later answers keep it.
  • Restate, don't assume. Anything defined far back in a compacted conversation, such as a constraint, a file name or a decision, is worth repeating once when it matters again.
  • Put instructions in messages. Lexika hands the summary to the model as reference, not as instructions, so a rule you want it to follow belongs in a message.
  • Save the durable things to the Brain. If the project's Brain is attached to the chat and you are the workspace Owner or an Admin, ask the assistant to remember a fact that matters beyond this thread, so it does not live only in an old message. A Member can read the Brain but not save to it.
  • Start a new conversation when the topic changes. A new conversation starts with nothing to summarize, and it reads the same memory once you attach the project's Brain to it. Keep a long thread to a single line of work.
  • Recent detail is safest. Compaction always covers the oldest part of a conversation and leaves at least the four newest messages out. The model reads the stretch you are actively working in verbatim.

In short

The rose marker shows where the model switches from reading a summary to reading your conversation word for word. Your messages are all still there. The model still works from what came before, as a summary above the seam and word for word below it, and if the summary gets something wrong, you can edit it.