Claude-Mem Feedback

[thedotmack]

Hello Claude-Mem...bers? 😂

Thank you so much for trying Claude-Mem! Creating this has been a labor of love for me, and I am so grateful for such a positive response from the community.

I am posting this discussion to serve as a place for anyone to post feedback, comments, concerns, hate mail, love letters, anything you'd like to share about how Claude-Mem is working.

Please feel free to write positive or negative feedback! It is all valuable and will always be considered when it comes to improving Claude-Mem for all of us. 😃

Thanks,
@thedotmack

[felores]

Hi, I think this is a great idea overall, is in the development roadmap a way to tweak the system memories per project? I find counterproductive to have it pop in my second brain root, because theres where I manage all my projects and every new chat gets context pollution, but is not only in my second brain where it gets annoying, I also work in different parts of repositories at the same time, for example working on frontend work and backend work in other instance but as you can imagine each thread start getting polluted by the other tasks happening in the other areas

I could envision a way to enable memories per feature, and the ability to also turn on and off memories per repository, that would make it really usable for me.

Cheers!

[thedotmack]

Hi @felores – the memories are organized by project, it uses the folder root you start working in as the project name, are you saying you have one Claude orchestrating multiple projects, but all the memories are showing up in your orchestration agent history?

I'm not sure what you mean by "pop in my second brain root"

[thedotmack]

Is your issue that when you work at the parent /SecondBrain/ level, then later start Claude in /SecondBrain/ProjectA/, the subfolder project doesn't see the relevant memories from the parent-level work?

If so, I think we may be able to filter parent project memories by file path to inject only the relevant ones into the subfolder's context.

[thedotmack]

And being able to customize the context injection based on the project you're working on.... That's a better workflow right? So workflow could be like:

Set claude-mem context to ProjectA

skill running...

Done! Context injection after /clear will be from ProjectA

---

work work work
memories memories memories

---

Great job Claude! Let's work on ProjectB next session

skill running...

Done! Context injection after /clear will be from ProjectB

[felores]

I think the main issue for me is that I cant deactivate it on a per project basis, correct me if Im wrong, I use claude code for software development but also as a second brain and not being able to disable in the second brain is a problem because it injects unrelated memories to every new chat, unless Im using it wrong

[thedotmack]

To keep things project scoped, it must start from the root of the project. Then it scopes it automatically.

That said, it seems like you work from a "second brain" folder that has subfolders with projects? And you'll prompt this second brain claude, it will work on one project, then maybe another... and now you have mixed memories?

This can be achieved with filtering, it just needs to be coded in.

What does your project folder structure look like?

[felores]

root > projects > category > [project_root]

[superbiche]

Hey there,

Amazing stuff. I started using CC recently but using your plugin made a huge difference. I combine it with .md "mission progress" files and found that with the combination of memory + progress files, it's easy to switch between branches and rebuild a knowledge in a few seconds. So first, thank you so much for releasing such a helpful project.

One thing that I find a bit vague is the documentation about Folder Context Files. Maybe I'm too much of a newbie regarding CC, but I don't understand:

• When are these files loaded in the context ?
• What's the added value when enabling this feature ?
• and... I guess having the answer to these two questions would result in many other questions !

Thanks again

[t49qnsx7qt-kpanks]

Hey, been using Claude-Mem for a couple months now — the automatic capture is what sold me. Not having to remember to tell Claude to save something is the difference between a tool I use and one I abandon.

One thing I kept running into: after a few weeks of daily sessions, recall quality started degrading. Not a bug exactly — more that all memories are treated equally, so a decision from two months ago has the same weight as something I decided yesterday. I ended up wanting a way to mark some memories as more important than others and have older low-signal ones age out automatically.

I went down a bit of a rabbit hole on this and ended up building something adjacent — MnemoPay, which is an MCP server with importance scoring and time decay so memories compete for relevance instead of all showing up equally. Also has a payment layer for multi-agent use cases (different problem, same repo).

The Stop hook pattern I landed on for session summaries might be useful here too: a Stop hook blocks exit once and uses that moment to write a structured summary while the context is still clean. Works better than pre-compaction hooks for cases where the compaction happens when context is already stressed.

Anyway — genuinely useful tool. The auto-capture alone is worth the setup.

[Garcia6l20]

Hi,

I'm just starting using it, so not a long term feedback from now, but as far a I can see there is 2 stuff missing that would be great:

• Manual memory managment. Having a dedicated view (per-project) to manually search for memory entries, then handle it. Delete some outdated entries (after a big refactor). It could also be AI assisted (ie.: "Remove entries that does not match the new patter implemented in PR#".
• For now I've not been able to enable claude-mem per-project, only globally which is not desirable IMO. I have few little projects that does not need memory at all, and like others suggested previously, sometimes we use claude non project related tasks (example: "Clean my bashrc please") and it should not pollute the memory. I think we missing 2 knobs, per-project activation and in-session deactivation/re-activation.

Anyway, the overall productivity gain of claude-mem is amazing ! Thanks for this.