A place for the work
Project context, tool use, memory, and recovery should not vanish every time a chat ends.
COGG9 research notebook
COGG9 is where I’m keeping track of the AI research and development I’m doing.
The work is split into three projects, but they are really pieces of one bigger attempt: an AI harness for long projects, a custom AI model that can grow over time, and a compression research lane for carrying meaning more efficiently inside machines.
It is early, messy in places, and still changing. That is why I’m documenting it.
Still early
Local hardware pressure
Shown honestly
What COGG9 is
Three projects, basically one larger attempt.
COGG9 is my home for keeping track of the AI research and development I’m doing.
HMS is the harness, the body and workspace around long AI projects. PlasticNodeLM is the model/core direction. WaveCodecLLM is the compression and signal research. They are split up now because the work needs rooms. Eventually they may fit together.
I’m not pitching this as a finished product. I’m showing the work while it is forming.
The projects
A body, a brain direction, and a signal lane.
That is the plain version. HMS gives AI work a place to live. PlasticNodeLM asks what the model/core could become. WaveCodecLLM asks how meaning might move inside the machine without wasting so much compute.
01AI harness
HMS
HMS is my AI harness for long-running work.
HMS is the harness I’m building around AI work, closer to the kind of workspace people understand from Hermes, OpenClaw, and tool-using AI systems, but focused on my own needs: project context, tool use, memory, recovery, and owning the work instead of losing it across scattered chats.
- AI Harness
- Project Memory
- Tool Use
- Long-Running Work
02AI model
PlasticNodeLM
PlasticNodeLM is my custom AI model direction.
PlasticNodeLM is my attempt at creating a custom AI model. The idea is not just a chatbot with tools bolted on. I’m researching systems that could let a model develop structure over time instead of staying frozen in one state.
- AI Model
- Model Growth
- Custom Architecture
- Local Hardware
WaveCodecLLM is my compression and signal research lane.
WaveCodecLLM is my attempt at tackling high compute and inefficient internal communication. Human words are useful at the edge, but inside the machine there may be better ways to carry signals, compress meaning, and still keep the work inspectable.
- Compression
- Machine Meaning
- Signal Paths
- Compute Pressure
What ties it together
I’m trying to give AI work a place to live, a model/core to grow from, and better signal paths inside the machine.
A core that can change
PlasticNodeLM asks whether a model can develop useful structure over time instead of staying in one frozen shape.
Less waste inside
WaveCodecLLM asks whether machines can carry meaning more efficiently while people still get inspectable results.
Normal machines matter
The work keeps pressure from local hardware in view. If it only works in fantasy compute, it is not grounded enough.
The work is happening on a real local machine, not magic cloud compute.
The current COGG9 test box is an Ubuntu workstation with a Ryzen 9 CPU, 64 GB system memory, and two Intel Arc Pro B70 GPUs with about 30 GB of VRAM each. It is the machine I use to keep local hardware pressure in view.
CPUAMD Ryzen 9 3900X, 12 cores / 24 threads.
GPUDual Intel Arc Pro B70 cards for XPU/local AI testing.
Memory64 GB system RAM, with local storage for experiments and artifacts.
Why local hardware matters
I want the work to stay close enough to inspect.
Normal hardware creates pressure. Memory, compute, recovery, and tool use all look different when you cannot pretend the machine is infinite.
- Own the data and workflow
- Keep local hardware pressure visible
- Use tools with review points
- Show the research without exposing the blueprint
Patreon
Follow the build while it is still forming.
The Patreon is for people who want the research notes, build logs, dead ends, and plain-English updates as the work matures. No finished platform pitch. Just careful work, shown honestly.
Contact
Contact page
COGG9 is still being built in public.
For research questions, archive notes, or contact about the work: Edward Greenwood, cogg9research@gmail.com.