I’ve launched a chat agent with access to only one specific project documentation to serve as a quick access point for project-related information. I was testing it when it randomly brought up a 1:1 meeting with my manager I had been preparing for 2 months prior - way before this chat agent existed. That note was auto-saved in the default chat agent’s memory, so it seems like chat agents share memory. This is bad, because it means they can cross-contaminate data from different projects - for example, if a chat agent for project X saves to memory that milestone 10 passed 79% of the deliverables, and I later ask project Y chat agent for the pass rate of MS10, it may pull info from their shared memory and give me project X numbers instead.
I searched through agent settings and existing documentation and I can’t find a memory filter anywhere.
Have I set this up wrong? Does the option exist somewhere? Or maybe there’s a known workaround? Ideally, memory could be either disabled or restricted per chat agent - otherwise, they may end up being unusable if someone works with more than one project.
First of all welcome to the Quick suite Community.
As per my understanding Memory is user-specific, not agent-specific**.** The memory system in Quick is designed to be personal to your individual account—it stores your preferences, context, and past interactions across all your conversations, regardless of which chat agent you’re using. Refer here
What it could means:
When you save information to memory (like your 1:1 meeting notes), it’s stored in your personal memory
This memory is accessible to any chat agent you interact with, including custom agents you create
Memory is not isolated per agent—all agents can access the same pool of your personal memories
I think You haven’t set anything up wrong**.** This is the current design of the memory system i could sense.
Unfortunately, there is no memory filter, disable option, or per-agent isolation setting available in the current version of Quick as far as i know.
Potential workarounds you could consider:
Provide explicit context in each conversation - Rather than relying on memory, explicitly attach relevant files or reference specific data sources for each project
Clear outdated memories periodically - You can review and manage your memories by going to Conversation History .
Just in case someone finds this, I did manage to get an acceptable result (though not perfect) by specifying in the context window in chat agent details that it should clearly mark what information it takes directly from documentation, what is an assumption, and what is taken from memory. I can just manually ignore the latter. Just telling it to not use memory did not work at all, so I guess this is the best I can get.