Claude “Compacting our conversation”: what it means, the fix

"Compacting our conversation so we can keep chatting" means your Claude chat has grown close to its context window, the amount of text Claude can hold in one conversation. Instead of stopping you, Claude summarizes the earlier messages and carries on from that summary. The fastest fix is to do nothing: let it finish, then restate any rule or number that matters before you continue.

If compaction fails or keeps coming back, move to a fresh chat with a short handover, or a Project for long-running work. Sources: Anthropic's help center, checked October 22, 2026.

TL;DR: Compaction is Claude shrinking the start of a long chat into a summary so the chat can keep going. Your history stays visible, but Claude now works from the summary, and long chats cost more usage. If it fails, start a new chat, paste a handover summary, and keep reference files in a Project.

Why you see it

Anthropic's help article How do usage and length limits work? explains the mechanism. Every chat has a fixed context window, and on paid plans the newest models support up to 1M tokens, others 500K or 200K. A share of that window is always reserved for Claude's reply.

When a conversation approaches that limit, Claude "summarizes earlier messages" so you can keep going. Anthropic calls this automatic context management and says you "may occasionally see that Claude is 'organizing its thoughts'" while it happens. The on-screen wording varies: people report "Compacting our conversation so we can keep chatting", and in Cowork, "Compressing our conversation so we can keep chatting."

Three things make it show up sooner:

• Big inputs. Long PDFs, pasted spreadsheets and images fill the window fast.
• Long back-and-forth. Every reply, including Claude's own long answers, stays in the window.
• Code execution being on. Anthropic says "code execution must be enabled for automatic context management." Without it, a long chat hits a hard length limit instead of compacting.

What compaction keeps and what it loses

Anthropic says "your full chat history is preserved so Claude can reference it even after summarization." You can still scroll up and read everything. What changes is what Claude actively works from: a condensed version of the early part of the chat, plus the recent messages in full.

Summaries keep the gist. Most likely to blur: exact figures, long quoted passages, and fussy instructions you gave once at the start ("never use bullet points", "keep it under 300 words"). Anthropic does not publish what the summary includes, so treat anything you said early as possibly softened.

How to fix it

The toggle path comes from Anthropic's file creation article. On mobile, tap your name in the sidebar to open Settings, then Capabilities. On Team and Enterprise, the owner controls it in Organization settings.

When compaction fails

Compaction has failed loudly at times. On GitHub, people using the claude.ai web, desktop and iOS apps reported "Compaction failed unexpectedly" in January 2026, with prompts that "revert to draft" afterwards (issue 18482). Others saw "Conversation compaction failed" and a chat stuck on "responding in background" (issue 20696).

A third string, "Exceeded max compactions per block", appeared in the Windows and macOS apps in February 2026 (issue 23313). Those threads were filed in the Claude Code tracker by mistake and closed without an Anthropic reply, so there is no official explanation for any of these three messages.

What the reporters found, which is a workaround and not a documented fix:

1. Reloading, logging out or switching devices did not rescue a broken chat in issue 20696.
2. If a reply was cut off, press Retry rather than typing "continue", which made the partial answer vanish for one reporter.
3. Start a new chat. It was the only recovery that worked reliably, and it is the one Anthropic itself recommends: "if you hit a length limit, you can start a new conversation."

If Claude says the chat can't be compacted any further, or stops accepting messages, the same applies: the work moves to a new chat.

Continue in a fresh chat without losing the thread

A new chat gets a fresh, empty context window. The trick is a handover. If the old chat still answers, send it this:

Write a handover note for a new chat. Include the goal, decisions made, key numbers, my formatting rules, and the next three steps. Keep it under 400 words.

Copy the note, open a new chat, paste it, and attach the one or two files the work depends on.

On paid plans there is a second route. Anthropic's chat search and memory article lets you ask "Let's continue where we left off with [project]." and Claude searches your past chats. Inside a Project, that search stays within the Project.

How to stop it happening again

• Put reference files in a Project. Anthropic says Projects use retrieval, "only loading relevant content into the context window", so a 40-page report does not sit in every message. Our Claude Projects guide walks through the setup.
• Start fresh before you hit the wall. If a chat has run all afternoon, a handover now beats a failed compaction later.
• Avoid giant first messages. Anthropic warns that "very large first messages" may still hit context limits, compaction or not.

Honest limits

Compaction is not free. Anthropic says "longer conversations that trigger automatic context management consume more of your usage limit", so a long compacted chat brings you closer to your usage cap. See how Claude's usage limits work.

There is no setting to turn compaction off while keeping code execution on, and no way to choose what the summary keeps. Anthropic documents no fix for the failure messages above. We did not reproduce them ourselves: everything in the failure section comes from user reports.

Questions people ask

Does Claude forget everything when it compacts?

No. Your full history stays in the chat and Claude can still reference it, per Anthropic. But Claude works mainly from a summary of the early messages, so exact wording, numbers and one-off instructions can fade. Restate anything critical after compaction.

Can I turn off compacting in Claude?

Not directly. Anthropic ties automatic context management to the code execution setting, and turning that off means a long chat stops at a length limit instead. The practical option is shorter chats and reference files kept in a Project.

Does compacting use up my Claude usage?

Yes. Anthropic says longer conversations that trigger automatic context management consume more of your usage limit, and suggests starting a new conversation if you are close to your limit in a long chat.

What does "Compaction failed unexpectedly" mean?

It means Claude tried to summarize your long chat and could not. Anthropic has not documented this message. Users who reported it found that only a new chat helped, ideally seeded with a handover summary copied from the old one.

Long chats are where most Claude surprises happen. The newsletter sends one short note when Anthropic changes a limit like this one.