Claude Compacting Our Conversation: What It Means (2026)

Claude compacting your conversation means it is summarizing the earlier part of a long chat so it fits the model's context window and you can keep going. Anthropic calls this automatic context management: it runs on paid plans with code execution on, does not count toward your usage limit, and keeps your full chat history (Anthropic Help Center, 2026). The summary keeps the gist, not every line.

This guide explains what compaction does on claude.ai, why it happens earlier on some models than others, what you lose and keep, and how to control it, including the /compact command in Claude Code and the compaction setting in the Claude API. It is based on Anthropic's Help Center and developer documentation as of September 2026. The last sections show how to see compaction coming with the free context window meter in AI Toolbox (formerly Claude Toolbox), a Chrome extension with 40,000+ users, a 4.7/5 Chrome Web Store rating, and a free plan.

By Adi Leviim, Co-Founder at Infi Developments. 7+ years building AI productivity tools. Last reviewed September 2026.

What "Compacting Our Conversation" Means in Claude

Compaction is Claude summarizing earlier messages when a conversation approaches the limit of its context window, the amount of text the model can work with at once. Anthropic's Help Center describes it this way: "When your conversation approaches the context window limit, Claude summarizes earlier messages to continue the conversation seamlessly" (Anthropic Help Center). The status people search for reads along the lines of "Compacting our conversation so we can keep chatting", and Anthropic's articles also describe the same moment as Claude "organizing its thoughts".

Before automatic context management, a long Claude chat simply stopped at the limit and you had to start over. Compaction is the reason a long chat can now continue "indefinitely in most cases", in Anthropic's words (Anthropic Help Center).

When Claude Compacts: Context Window Sizes by Plan

Compaction starts when the chat nears the context window of the model you are using, so bigger windows compact later. On paid plans in claude.ai chat, Anthropic lists these windows as of September 2026:

A portion of every window is reserved for Claude's reply, so the usable conversation is slightly smaller than the headline number. Tools and connectors are token-intensive, so a chat with many connectors, long files, or extended thinking reaches the threshold sooner. The Claude context window guide converts these token counts into pages and words.

What Compaction Keeps and What It Can Lose

Compaction keeps the thread of the conversation and drops the verbatim detail of older turns, because the model now works from a summary of them. The chat still looks complete on your screen, and Anthropic says Claude can reference the preserved history, but the model's working context for the older part is the summary.

• Written to be kept: the information needed to continue the task, which is what Anthropic's default summarization prompt asks for; in Claude Code, Anthropic says your requests and key code snippets are preserved.
• Can blur: detailed instructions from early in the conversation, which Anthropic says may be lost, along with exact figures and long quoted passages.
• Worth restating after a compaction: hard constraints, such as a word limit, a style rule, or a number the answer must use.

Anthropic's developer docs describe the same trade-off from the other side: as a conversation grows, response quality degrades, so compaction replaces older content with a summary to keep the active context small (Claude Platform Docs).

How to Handle Compaction on claude.ai

claude.ai has no per-chat switch for compaction; it depends on the code execution setting. You can still decide what survives it and when a fresh chat is the better choice.

1. Pin the essentials in one message. Before a long session, or right after a compaction, send a short list of the facts and rules Claude must keep.
2. Move reference material into a project. Projects use retrieval, so Claude loads only the relevant parts of your files instead of carrying everything in the chat (Claude projects guide).
3. Trim what fills the window. Turn off connectors and tools you are not using, and keep project instructions short, as Anthropic recommends.
4. Start a fresh chat for a new task. Paste a short summary of where you are, rather than letting an unrelated task compact the old one.
5. Save the conversation first if exact wording matters later. Export it to a file (Claude export and bulk delete).

If you see a hard stop instead of compaction, check Settings, then Capabilities: Anthropic notes that code execution must be enabled for automatic context management. That setting is also where you turn on file creation (Anthropic Help Center). Hitting a usage limit is a different thing with a different fix, covered in the Claude usage limits guide.

See compaction coming instead of finding out afterwards. AI Toolbox's context window meter shows how full your Claude chat is, live inside claude.ai, with a hint at 80% and a prompt to continue in a fresh chat at 95%. The meter is free. 40,000+ users, 4.7/5 on the Chrome Web Store.

Add AI Toolbox to Chrome, free ->

Compacting in Claude Code: /compact and Auto-Compact

Claude Code compacts automatically as you approach the context limit, and you can run the same step yourself with /compact. The automatic pass summarizes the conversation history so a full context window does not end your session (Claude Code Docs). The commands Anthropic documents:

After compaction, Claude Code re-injects the project-root CLAUDE.md and auto memory from disk and re-reads up to five of the files it recently read or edited, most recently modified first. Background commands and subagents keep running. That is why Anthropic recommends putting persistent rules in CLAUDE.md rather than relying on conversation history, and adding a "Compact Instructions" section to CLAUDE.md to control what compaction preserves (Claude Code Docs). If one huge file or tool output refills the window right after each summary, Claude Code stops auto-compacting after a few attempts and shows an error instead of looping.

Compaction in the Claude API

Developers can turn on server-side compaction in the Messages API so long conversations summarize themselves at a threshold you choose. Add the compact_20260112 strategy to context_management.edits; the default trigger is 150,000 input tokens and the minimum is 50,000 (Claude Platform Docs).

• The API returns a compaction block with the summary, and ignores everything before that block on later requests.
• Custom instructions replace the default summarization prompt entirely, so state what the summary must retain.
• Compaction adds a sampling step that counts toward rate limits and billing, and the same model does the summarizing.
• The compact-2026-09-04 beta lets you request a summary on demand instead of at a threshold.

Native Claude vs AI Toolbox: Managing Long Chats

Claude handles the limit for you once you reach it; AI Toolbox's Claude module shows you where you stand before that and keeps what you need outside the chat.

The meter is calculated in your browser from the conversation cached on your device and sends nothing. Continue in a fresh chat is different: it sends the conversation text to AI Toolbox's summarizer to write the summary, so it is off unless you choose it. The context window meter guide covers both in detail.

Who Runs Into Compaction Most

• Developers in Claude Code: long debugging sessions compact often. Put project rules in CLAUDE.md so they are re-injected after every compaction.
• Analysts working through large files: uploads fill the window fast. Keep source files in a project so Claude retrieves only what a question needs.
• Writers drafting long documents: style rules from early in the chat can blur. Restate them after a compaction, or keep them in project instructions.
• Researchers running one chat for weeks: export the conversation at milestones so the exact wording is saved outside Claude (search Claude history).

Limitations and What Native Claude Handles Well

Automatic context management is a real improvement: before it, a long chat ended at the limit. Claude now keeps going, preserves your history, and does not charge the compaction against your usage limit. It is not available when code execution is off, rare cases like a very large first message can still hit the limit, and no tool can make a summary as exact as the original messages. AI Toolbox does not change how Claude compacts; it shows how full the window is and keeps copies you control.

Frequently Asked Questions

What does compacting our conversation mean in Claude?

It means Claude is summarizing the earlier part of a long conversation because the chat is close to its context window limit. The summary replaces the older messages in Claude's working context so you can keep chatting. Your full history stays visible, and Anthropic says compaction does not count toward your usage limit.

Does compacting delete my Claude conversation?

No. Anthropic says your full chat history is preserved so Claude can reference it even after summarization. What changes is Claude's working context: older messages are represented by a summary, so exact details from early in the chat can be less precise.

How do I stop Claude from compacting?

On claude.ai there is no switch for a single chat; it runs automatically on paid plans with code execution enabled, and turning code execution off means long chats stop at the limit instead. You can delay it by keeping chats focused, moving files into a project, and turning off unused connectors. In Claude Code, /autocompact sets how full the window gets before it runs.

Why is Claude compacting so often?

The chat is filling the context window quickly. Large uploads, many tools or connectors, and long outputs all use tokens, and models with a 200K window reach the limit sooner than 500K or 1M models. Moving reference files into a project and starting new chats for new tasks reduces how often it happens.

Does compaction count toward my Claude usage limit?

No. Anthropic's Help Center states that automatic context management does not count toward your usage limit. Usage limits and length limits are separate: usage limits cap how much you use Claude over time, and length limits cap how long a single chat can grow.

What does /compact do in Claude Code?

/compact replaces the conversation with a structured summary right away, the same step the automatic pass runs near the limit. Add instructions, such as /compact focus on the auth bug fix, to choose what the summary keeps. CLAUDE.md and auto memory are re-injected from disk afterwards.

How can I see how full my Claude chat is?

claude.ai does not show context usage. AI Toolbox's free context window meter shows an estimate live in the chat, with a hint at 80% and a prompt to continue in a fresh chat at 95%. In Claude Code, run /context for a live breakdown.

Bottom Line

"Compacting our conversation" is Claude keeping a long chat alive by summarizing its older part. It is automatic on paid plans with code execution on, it does not use your usage limit, and your history stays. Restate what must stay exact, keep reference files in projects, start fresh chats for new tasks, and watch the meter so compaction never takes you by surprise.

Related reading: Claude context window and token limits, Claude context mentions, Claude folders, and Claude vs ChatGPT.

Sources

• Anthropic Help Center, How do usage and length limits work?
• Anthropic Help Center, How large is the context window on paid Claude plans?
• Anthropic Help Center, Create and edit files with Claude
• Claude Code Docs, Explore the context window
• Claude Code Docs, How Claude Code works
• Claude Platform Docs, Compaction

Last updated: September 21, 2026