You are mid-task and Claude stops: "This conversation is too long to continue. Start a new chat, or remove some tools to free up space."

Read the second half of that sentence again. Anthropic put "remove some tools" in the error text for a reason. On a lot of accounts the conversation is not the thing filling the budget, and starting a new chat drops you straight back into the same wall a few messages later.

Quick answer: Open "Search and tools" and turn off connectors you are not using right now, then enable code execution so Claude summarizes the thread automatically instead of stopping. In Claude Code, run /compact to continue or /clear to start a new task. In Claude Desktop there is no compact option at all, so move long work to the web app.

First, check which wall you actually hit

Claude produces three different limits that people report as one problem. They have different fixes, so spend five seconds identifying yours.

	Conversation length limit	Message size limit	Usage limit
The message	"This conversation is too long to continue"	"Your message will exceed the length limit for this chat"	You have used your allowance
What it means	The whole thread filled its context window	This one message plus its attachments does not fit in what is left	Your account used its allowance across every conversation
Scope	This one thread	This one message	Your whole account
Does a new chat help?	Yes	Sometimes	No
Does waiting help?	No, never	No	Yes

If you are reading "too long to continue," the rest of this guide applies. If you are reading "exceed the length limit for this chat," Claude is telling you to attach fewer or smaller files, and that alone usually gets the message through.

Usage limits are a separate system worth keeping straight in your head. They govern how much you can use Claude across every conversation and they come back with time. Length limits belong to a single thread and never come back.

The chat that is already too long on message one

Start here, because this is the version of the problem with the most dramatic fix.

A Claude user reported hitting "conversation is too long" on brand new chats at roughly 16k of a 200k window. That is about 8 percent. The report traces the cause to project instructions and connected tools, which were eating the budget before a single message was typed.

This is not a bug so much as an accounting surprise. Every tool and connector you enable ships its definition into the context on every single turn, whether you call it or not. Anthropic's own help documentation describes tools and connectors as token-intensive and recommends temporarily disabling the ones you do not need. Project instructions and project files behave the same way.

So the diagnostic is simple. If a fresh chat dies early, your setup is the problem, not your conversation.

Fix it: reclaim the budget before you restart the thread

Every step below comes from Anthropic's guidance on usage and length limits. Do them in this order, because the first two do most of the work.

1. Turn off connectors you are not using. In the "Search and tools" settings, temporarily disable non-critical tools and connectors: web search, Research, MCP connectors. Turn them back on for the turns that actually need them. This is the fix the error message itself is pointing at.
2. Enable code execution. On paid plans this is what switches on Claude's automatic context management. With it on, Claude summarizes earlier messages as the thread approaches the limit and continues instead of stopping, and your full chat history is preserved so it can still reference it. If you see Claude "organizing its thoughts" during a long conversation, that is the mechanism running.
3. Shorten your project instructions and delete unused project files. Both are re-sent on every turn. Instructions you wrote for a task you finished last month are still costing you space today.
4. Turn extended thinking off, or lower the effort level. Anthropic lists both as ways to free up space when you are running tight.
5. Move reference material into a Project. Projects use retrieval, so only the relevant parts of your documents load for a given turn rather than all of them at once.

Two caveats from that same help page, both of which most write-ups skip. Automatic context management requires code execution to be enabled, so if you never turned it on, you never had it. And rare edge cases, such as a very large first message, can still hit the context limit even with everything configured correctly.

There is also a price. Anthropic notes that conversations long enough to trigger automatic context management consume more of your usage limit, so the feature that saves your thread will move you toward the other kind of limit faster.

How much room you had to begin with depends on the model. On paid plans, Anthropic says the newest models support up to 1M tokens while others sit at 500K or 200K, with a portion always reserved for the response.

What happens at the limit depends on where you are running Claude

This is the part that makes advice about Claude so inconsistent. The same account behaves differently in three places.

Surface	Behavior at the limit	Your best move
Claude web or mobile, paid plan, code execution on	Summarizes earlier messages automatically and keeps going, full history preserved	Watch for "organizing its thoughts" and keep working
Claude web, code execution off	Hard stop with the "too long" error	Enable code execution, then prune tools
Claude Desktop	Reported hard wall with no compact option and no warning	Prune tools, and move long work to the web app
Claude Code (CLI)	Auto-compacts near the limit, and you can compact on demand	/compact to continue, /clear for a new task

Claude Code gives you the most control, because compaction is a command rather than a background process:

• /compact summarizes the history so far and frees space. Use it when you are continuing a long task.
• /clear wipes the conversation while keeping your CLAUDE.md and project files. Use it when you are starting a new task. It cannot be undone.
• /rewind, or a double tap of Esc, jumps back to an earlier message and drops everything after it.

Anthropic's rule of thumb is worth memorizing: /clear when you start a new task, /compact when you continue a long one.

Claude Desktop is the outlier, and you should know this before you plan a working session around it. A GitHub issue filed as SYS-029 reports Desktop hitting a hard wall at the context limit with no compact option, no autocompact and no warning beforehand. Commenters confirmed the same behavior across versions 1.2278.0, 1.2581.0 and 1.3036.0.

Why long threads get worse before they break

Sometimes there is no error at all, just decay. You set a constraint at the start and Claude quietly stops honoring it.

Anthropic describes the cause directly in its guidance on long sessions: "context rot is the observation that model performance degrades as context grows because attention gets spread across more tokens, and older, irrelevant content starts to distract from the current task."

Context rot is why a bigger window is not automatically a better experience. It is also why compaction is riskier than it looks. Anthropic notes that bad compacts happen when the model cannot predict the direction of your work, and puts the problem bluntly: "due to context rot, the model is at its least intelligent point when compacting."

Three habits help more than any setting:

• Compact early, on purpose. A summary written with room to think beats one written at the ceiling.
• State the direction before you compact. Tell Claude what you are about to do next, so the summary keeps what that work needs.
• Copy critical values verbatim. Summaries paraphrase, and paraphrasing corrupts exact numbers, names and code.

If you want the full mechanics of why every assistant hits this, we wrote it up in why AI assistants forget your conversation.

Stop having this problem

Everything above is real and it works. It is also you doing Claude's filing: auditing your connectors, watching a budget you cannot see, remembering to compact before the model gets too dull to compact well.

The root cause is structural. Your context lives inside a thread, inside one app, on one device. The thread is the unit of memory, so when the thread ends the memory ends with it. Automatic context management is genuinely good, and it still only extends the life of a single thread.

folk is built the other way around. It compacts long conversations too, that part is not unusual. What is different is that its memory is not scoped to a thread at all. It lives with you in iMessage and Telegram, so there is no conversation to reach the end of and nothing to re-paste when one dies. What you told it last month is still there this month. Most competing fixes for this problem are browser extensions bolted onto someone else's chat window, which is fine at a desk and useless when you are texting from a train.

If you already have context in Claude worth keeping, you do not have to retype it. You can import your ChatGPT and Claude memory in a single paste.

Hitting this on a different assistant? See the guides for ChatGPT and Gemini, or the overview of why every AI assistant forgets.