Watched Some of Your YouTubes...Nice Work.. Have a Question...

[aiartalchemy]

So, Can you run with a local Ai model like Qwen3.5 or any other model that will run on local computer?

A few thoughts;

1. I think as AI progresses, security and privacy are going to become huge issues. Use an API and share all your data with OPENAI, Google, Grok or Anthropic or use CoPilot and everything you do is logged and saved my Microsoft. No Thank You. It would be nice to have a home assistant that is completely private even if it has fraction of capabilities of Claude.
2. I have some idea how could manage the memory file so you don't end eventually overwhelming the context size so large that it loses personality. If you are interested. You don't want that memory context size getting too large. As it grows it just going to average out, Things will got lost and it's costing extra money because of input token costs. You want some sort of system that only inject relevant memories to the current last prompt to save on context size and input tokens. Plus you need some hierarchal system of memories so that really important memories get injected often while minor unimportant memories even get forgotten or erases or simply relegated to an unimportant hierarchy. Humans work this way. If you hear a song, we remember things related to the song only, what we ate for dinner last night isn't injected into our short term memory because there is no contextual relationship.
3. Agentic Ai with full control of my computer, accounts and access to the internet seems like a security risk. Also allowing the Ai to rewrite it owns prompts seems potentially problematic too. I suspect there will be website poising via prompt injection going on by trolls and hackers. Am I correct that you can run this and simply turn off various tool to prevent the ai from Accessing certain things?
4. I could also probably help you design a personality system that would simulate something closer to human behavior but it would act more like AI companion/Pet/Waifu. Like actually give it some needs and emotional responses when those needs aren't met or are met which modify it.

[ddxfish]

1- Hell yeah! You belong here. Self hosted.
2- We are experimenting with a new technique. Yes, more human is all I can say. I have been trying a few things for a year now. I think I found a technique that works. It's a simple drop in for our current memory. Mine is running it for maybe a week and yeah junk drops out and important stuff is boosted. You can probably imagine how we are doing it. And you might see it this month if it goes well in this test...
3- You have to battle my long series of popups containing a series of increasingly complex riddles to use those features.
4- That is a niche I won't tackle for 6-12 months. Want to make a plugin? You'd be the first. Have it toggle happy3 > happy4 when the LLM gives positive happy valence based on a second pass to the LLM just for emotional values. Tell me if you need more places to tap into the system from a plugin. I'll help you get there. Plugin store is new since last night on the website.

@aiartalchemy welcome, good to have you here