PLEASE use on non critical data to test!

[milla-jovovich]

I just want to remind everyone, this IS a dev tool, so PLEASE remember to test on no critical data first!

[fuzzymoomoo]

@milla-jovovich Where is the fun in that :P?
I have it sitting smack bang in the middle of everything I am doing at the moment. The initial release version and it's doing a great job, about to upgrade. I'm very curious so see the difference between the initial release and now.

[GusFromSpace]

i use aaak to speed up search, but i dont use it for anything else because lossy......btw, works well with https://github.com/Howtoimagine/The_Kaleidoscope_Geometric_Compressor

[MrSuddenJoy]

@milla-jovovich I (and I guess some others too) are still waiting for you reply in certain issues.

[mr-gl00m]

Oh God. Were people not doing that?!

[PapaBearDoes]

You're not my real mom!

[pedrolucas167]

That's fair but honestly this is exactly how real-world testing happens.

Started with “non-critical data” → suddenly it's running in production

Curious to see how stable it gets after a few iterations.

[MrSuddenJoy]

@pedrolucas167 production does not necesarily equal critical data.

[sync0x]

I find the two tiered memory system, immediate "short-term" journalling to MD files and routine patrols for the "long-term" with the ONNX enhanced indexer, work wonders for transporting context from site to site. That is to say I can have relevant context from accomplishments transported between development and production hosts by synchronising the folders storing the database file and the MD journals.

I can choose to delegate indexing to my development host which has a suitable GPU with which to run the ONNX component, and those long term memories are portable between hosts. On the production system, the agent retains foreknowledge of my latest developments via short-term MD's, and my expectations in regard to testing the stability of those new features.

This is a handy faculty to have when hitting the limits of one's own cognitive capacity to comprehend the inner workings of their own projects. Now what I mean by that, is by keeping the complexity of my work regulated to being just a touch outside my own ability to visualise the entire system in operation all at once, I am at my most efficient in terms of maintaining the balance between new feature development and keeping existing features stable in production.

Of course, my use case affords me the luxury of near certainty that I will seldom attempt to use both my development and production hosts for creation or maintenance work simultaneously. Any time I were to do this, I would likely have to resolve conflicting updates to files within my file syncronisation utility (unison in my case, but Meld/Winmerge are my fallbacks). Only the MD files can have conflicts; I must never allow the SQlite database to be updated (before syncronisation) in two places!

As has been suggested elsewhere, integration with a network aware database is essential for true multiuser or even multisession workflows but even in such workflows, only one host really needs to run the indexer for long term retrieval. It need therefore not be a top priority to find a replacement for the SQlite backed ChromaDB. The challenge is in short-term memory storage and retrieval; to migrate from MD files committed through regular file IO into network aware database transactions handling essentially the same service, but with the database's far better record locking and concurrency than a passively coordinated network file system offers.

Inter-host communication would be essential. The longer the delay in informing other nodes of updates, the greater the difficulty in resolving conflicts will be. AI also adds its own unique dichotomy to this scenario in that it makes the automatic conflict resolution it is capable of handling, look easier than it actually is. It is only possible to appreciate what I am alluding to when more than one user with diverging interests is working with the same synchronised project folder.

For now, we should consider mempalace to be single-user ready today, and for that it works well with our favorite file level synchronisation tooling.

[Synchlaire]

@milla-jovovich yes mommy