Skip to main content

Documentation Index

Fetch the complete documentation index at: https://help.mem.ai/llms.txt

Use this file to discover all available pages before exploring further.

​
Overview

Mem Chat is powered by large language models (LLMs) — the same technology behind ChatGPT, Claude, and Gemini. Like all LLMs, Mem Chat works within a context window: a limit on how much information it can hold at once during a conversation. This article explains what that means, when you might run into it, and what you can do about it.

​
How Mem Chat uses context

When you ask Mem something, it doesn’t search your entire workspace all at once. Instead, it pulls in the most relevant information — notes, collections, or attachments you’ve referenced — and uses that as context to answer your question. The more notes you reference, the more data gets pulled into context. And the longer your conversation goes, the more context accumulates. Once the context window is full, Chat may start to behave less reliably — giving incomplete answers, losing track of earlier parts of the conversation, or stopping work mid-task. This isn’t a bug. It’s how all LLMs work — and it affects ChatGPT, Claude, and every other AI chat tool in the same way.

​
When you’re more likely to hit the limit

A few patterns are more likely to fill up your context window quickly:
  • Referencing large collections. If you @mention a collection with hundreds of notes, Mem will try to pull in content from those notes to answer your question. A collection with 200 notes can use up a significant portion of your context before you’ve even had a back-and-forth.
  • Very long conversations. The longer your chat session, the more history accumulates in context. This is less of a concern for short, focused questions — but if you’ve been going back and forth for a while, it can add up.
  • Asking Chat to perform large bulk operations. Tasks like “organize all my notes” or “clean up everything in my workspace” require Chat to work through a large number of notes. Mem processes these in batches (around 50 notes at a time) and will check in with you before continuing — this is intentional, so you can review what it’s done so far and decide whether to keep going.

​
What to do when you hit the limit

Start a fresh chat. This is often the simplest fix. A new chat starts with a clean context, so Chat can focus on your question without carrying the weight of a long previous conversation. Don’t try to “rescue” a conversation that’s gone off the rails — start fresh and be more specific. Ask by topic, not by collection. Instead of attaching a large collection to the chat context (either by mentioning it or adding it as context manually in the chat via the ”+” menu), try describing what you’re looking for in plain language. For example:
  • ❌ “Look through my User Interviews collection and find themes”
  • ✅ “What themes came up in my customer interviews from last month?”
The second approach lets Mem search for what’s relevant rather than pulling your entire collection into context. Be more specific and add a timeframe. The more precisely you describe what you need, the less Mem has to pull in. Adding a timeframe is one of the easiest ways to narrow things down:
  • ❌ “Find all my resumes”
  • ✅ “Find resumes I saved in the last two weeks”
Switch to a model with a larger context window. Pro users can choose their AI model in chat. Different models handle their context windows differently. If you frequently work with large collections or long sessions, you can try upgrading to Pro and selecting a different model to see if it handles your task better.

​
About the “Would you like me to continue?” prompt

If you ask Mem to perform a large operation — like organizing or cleaning up hundreds of notes — you’ll see a message like:
“I’ve reached the limit for how much I can do in one turn. Would you like me to continue with the remaining work?”
This is intentional. Mem pauses to check in so you can:
  • Verify the work looks right before it continues.
  • Stop if something seems off — especially useful for operations like deleting duplicates or reorganizing collections, which can be hard to undo.
To check what Mem has done so far, click Show thought process in the chat. You’ll be able to inspect a more detailed log of the thought process that Mem took as it was carrying out the task. We know this can feel repetitive for large tasks. We’re exploring ways to let you tell Mem to continue without confirming every batch — but for now, we’ve kept the confirmation step as a safety measure for you and your data.

​
FAQs

Why did Chat stop working mid-conversation?
Your context window likely filled up. Start a fresh chat, be more specific about what you’re looking for, and avoid referencing large collections all at once. Can I undo something Chat did?
Deleted notes go to your trash, so you can recover them if needed. For other operations (reorganizing, renaming, editing), Mem doesn’t currently have a full undo from chat — which is one reason we pause and ask you to confirm during large bulk operations. For changes that Chat makes within your Mem workspace, you’ll often be able to see what it did in your All Notes and All Collections views, with Last modified as the sorting method. Will this get better over time?
Yes. The team is actively working on improvements to how Chat handles large workspaces and long-running tasks. In the meantime, the tips above should help you get the most out of Chat within current limits.