Auto-SummaryJakeoncid419 shares his journey over the past two years, shifting from mushroom growing to building Aether, an AI operating system designed to go beyond simple chatbots by integrating multiple models, memory systems, reasoning layers, and specialized tools into a cohesive framework. He outlines key components like the Primary Reasoning Stack, Mini Operating Core, and Guardian systems, emphasizing the importance of system design over individual models. Aether aims to enable AI to reason, remember, interact with software, and handle failures, with a focus on local capabilities and modularity. The project is still evolving, with ongoing efforts to unify its components into a more coherent system. Jakeoncid419 invites others working on similar AI systems to connect, share insights, and discuss challenges, framing the post as a project journal rather than a sales pitch.
I figured it was probably time I posted what I’ve actually been doing the last couple years instead of just occasionally showing up and posting old grows like some washed-up has-been 😂
I basically went down an AI rabbit hole and never came back.
What started as experimenting with local language models turned into a much bigger project I call Aether. At this point it’s less “chatbot” and more an attempt to build an AI operating system made up of multiple models, memory systems, reasoning layers, computer-use systems, media systems, monitoring, recovery, and a bunch of specialized tooling all working together.
I’m not posting this as a sales pitch. I want to use this thread as an ongoing project journal, show what I’ve actually built, document what works and what doesn’t, and hopefully connect with anyone else here working on local AI, agents, automation, model orchestration, computer vision, robotics, or related systems.
What Aether actually is
At a high level, Aether is my attempt to build a real AI system instead of just another prompt wrapper or chatbot.
The easiest way to describe it is: it’s becoming an AI operating system.
The language models are only one layer. The real goal is to have a system that can reason, remember, route work, use tools, interact with software, validate results, recover from failures, and eventually coordinate a lot of different specialized functions the same way an operating system coordinates programs and hardware.
I got interested in local models first, but the deeper I went the more obvious it became that the important part wasn’t just “which model is smartest,” it was the system around the models.
Major parts of the system so far
1. PRS — Primary Reasoning Stack
This is basically the executive layer. It helps decide what kind of reasoning path a task needs, what subsystem or model should handle it, and how work gets routed.
2. MOC — Mini Operating Core
This is the part that deals more with memory, runtime state, truth tracking, context assembly, permissions, routing, synchronization, and keeping the whole system from turning into spaghetti.
3. Hands & Eyes
This is the computer-use side. OCR, vision, UI understanding, action planning, interacting with apps/sites, checking whether actions worked, retrying when they didn’t, etc. Basically teaching the system to operate software instead of just talking about software.
4. Guardian / FailureMesh
One of the big things I learned is that AI systems fail in weird ways constantly, so I started building governance/monitoring/recovery logic around that. Instead of just treating mistakes as random bad runs, I’ve been trying to capture failures, figure out why they happened, and turn them into reusable guards so the system gradually gets more resilient.
5. VideoForge
This is the media side: image generation, enhancement, voice, animation, editing, rendering, packaging, and related automation. Long-term I want Aether to be able to help produce real content pipelines, not just text.
6. Model and tool layer
Aether isn’t tied to one model. I’ve been building it around a more modular idea where local models, cloud models, scripts, APIs, databases, media tools, automation frameworks, and other services can all plug into the same overall system.
What’s actually been done so far
A lot of the last 2 years has been building pieces, breaking pieces, rebuilding them, and slowly figuring out what the architecture really needed to be.
So far I’ve worked on things like:
- local model workflows
- multi-model orchestration
- persistent memory and retrieval systems
- structured context assembly
- agent/tool routing
- validation and self-checking logic
- runtime monitoring and recovery
- computer-use experiments
- media-generation pipelines
- system modularization and governance
Some parts are much more mature than others. Some things have worked surprisingly well. Some things completely failed and forced redesigns. Honestly that’s been a huge part of the process — not just building features, but learning what doesn’t work when you try to scale from “AI assistant” into “AI system.”
Current direction
Right now the bigger goal is to pull all these pieces into a more unified architecture so Aether becomes less like a loose collection of experiments and more like a true coherent system.
I’m especially interested in:
- better orchestration between subsystems
- stronger memory / context handling
- safer and more reliable computer-use
- stronger monitoring and recovery
- better local-first capabilities
- eventually distributed / multi-node setups
- media and automation workflows that produce useful real-world outputs
Hardware / setup
I’ll post some hardware pics too since the local side matters a lot to this project. Aether has been heavily shaped by the fact that I want serious local capability rather than depending entirely on outside services. So hardware, inference, VRAM constraints, model size tradeoffs, routing decisions, and performance bottlenecks all matter in a very real way.
[You can insert hardware pics here once you upload them.]
Why I’m posting this here
A few reasons:
1. This is genuinely what I’ve been doing instead of being active here the way I used to be.
2. I want to use this thread as a project journal and progress log.
3. I’d like to connect with anyone here working on local AI, agents, automation, computer vision, robotics, model orchestration, or related systems.
4. I’m really interested in comparing architectures, lessons learned, failures, and approaches with other builders.
I’m not here to dump all my internals or shoot myself in the foot by posting every secret sauce detail, obviously. But I’m happy to talk about the project at a meaningful level, show progress, discuss architecture, and document the journey.
If anyone else here is building in this space, especially local-first or systems-level AI stuff, I’d definitely like to hear about it.
|
Threaded
Previous
Index
Next
North Spore Mushroom Grow Kits & Cultivation Supplies 
Bulk Cannabis Seeds 
All-In-One Bags That Don't Suck 
Spores for European Microscopy
Jakeoncid419 
Edit 
Reply 
1 2 all )
therewatchingme
MetaShroom
Lana
Rainbow