Good idea. Though I was thinking to rebuild the memory layer from ground up. Currently it’s quite basic. Do you find it useful? Or do you prefer the memory to work similarly to ChatGPT?
David
•
Apr 30
The BoltAI memory is basic so a little more advanced options would be nice… and i fully support you rebuilding it…
some options i would like to see:
cross platform memory (PC, Mobile,…)
memory per project and global
option for AI to automatically save memory or only manually (ask him to save something to memory)
browse, search and edit memory interface
specific topic dynamic memory - bucket (like save what i ate each time i tell you to food bucket) so you can recall when i ask you anything about it next time what should i eat today...
probably a lot more :D stuff but all above would be awesome!
Matt
•
May 6
I don’t use the memory on Bolt as is, but I think there's a real opportunity here to create something which is quite distinct from the way ChatGPT does it, that works for power users and gives the user a lot more control over the memories.
Specifically (and repeating several of David’s ideas here):
- Per project memory. We all use AI for such a variety of topics that global memories don’t make a lot of sense/aren’t good for privacy. - Memory settings per project. Memory on/off. Autosave memories on/off. Prompt for memory generation. Length of memory to generate per chat (absolute number of characters or % of the chat being summarised).
- Button in the chat interface to manually save a memory of an individual response or of the whole chat (in the latter case if memory autosave isn’t enabled). - Memory management (view, search, edit) as suggested, but specifically with the ability to group memories, attribute them to different projects in bulk, and manually add memories that don’t originate in Bolt or weren’t saved at the time.
My ideal implementation of this would put memories in a subfolder of the synced Obsidian project folder from which context documents are already being pulled, but I’m guessing there’s quite a few reasons why you’ll never implement it that way…
But seriously, this is a big opportunity to take something that’s a weakness of Bolt compared to the big platforms, and turn it into something that’s original and powerful.
Log in to comment and vote
Comments3
Daniel Nguyen
Apr 30
Good idea. Though I was thinking to rebuild the memory layer from ground up. Currently it’s quite basic. Do you find it useful? Or do you prefer the memory to work similarly to ChatGPT?
David
Apr 30
The BoltAI memory is basic so a little more advanced options would be nice… and i fully support you rebuilding it…
some options i would like to see:
cross platform memory (PC, Mobile,…)
memory per project and global
option for AI to automatically save memory or only manually (ask him to save something to memory)
browse, search and edit memory interface
specific topic dynamic memory - bucket (like save what i ate each time i tell you to food bucket) so you can recall when i ask you anything about it next time what should i eat today...
probably a lot more :D stuff but all above would be awesome!
Matt
May 6
I don’t use the memory on Bolt as is, but I think there's a real opportunity here to create something which is quite distinct from the way ChatGPT does it, that works for power users and gives the user a lot more control over the memories.
Specifically (and repeating several of David’s ideas here):
- Per project memory. We all use AI for such a variety of topics that global memories don’t make a lot of sense/aren’t good for privacy.
- Memory settings per project. Memory on/off. Autosave memories on/off. Prompt for memory generation. Length of memory to generate per chat (absolute number of characters or % of the chat being summarised).
- Button in the chat interface to manually save a memory of an individual response or of the whole chat (in the latter case if memory autosave isn’t enabled).
- Memory management (view, search, edit) as suggested, but specifically with the ability to group memories, attribute them to different projects in bulk, and manually add memories that don’t originate in Bolt or weren’t saved at the time.
My ideal implementation of this would put memories in a subfolder of the synced Obsidian project folder from which context documents are already being pulled, but I’m guessing there’s quite a few reasons why you’ll never implement it that way…
But seriously, this is a big opportunity to take something that’s a weakness of Bolt compared to the big platforms, and turn it into something that’s original and powerful.