Memory from birth
How should continuity be considered from the first design step, not bolted on later?
COGG9 machine core
Local-first AI architecture with memory, structure, and human control.
COGG9 is an independent AI research studio exploring HMS, PlasticNodeLM, and memory-aware systems designed around visible boundaries and human guidance.
Research-grade
Human-guided
Local-machine focused
What COGG9 is
A public workshop for careful AI system design.
COGG9 documents experimental AI architecture from a builder’s point of view: local-first systems, memory from birth, permissioned tool use, model structure tests, and human-machine safety layers.
The work asks practical questions in public: what should an AI system remember, when should it ask permission, how should project state stay visible, and how can a person stay in control while still getting useful help from the machine?
Project pillars
Two core paths, one research direction.
HMS is the workspace around AI-assisted work. PlasticNodeLM is the architecture research path inside the machine.
01AI harness
HMS
The memory harness around long-running AI work.
HMS is the memory harness I’m building around AI work — a place for project context, decisions, open questions, and recovery points to live beyond a single chat.
- AI Harness
- Human-Like Memory
- Project Continuity
- Long-Running Work
02Growing AI core
PlasticNodeLM
A model architecture designed to grow, adapt, and eventually work through HMS as its body.
PlasticNodeLM is the growing AI core COGG9 is researching — a model architecture designed around development, continuity, adaptation, and eventually working through HMS as its body.
- Growing AI Core
- Model Organism
- Adaptive Structure
- Core Continuity
- HMS Body
The space between human words and machine meaning.
WaveCodecLLM explores how machines might carry meaning more efficiently behind the scenes while still speaking clearly to people.
Explore the space between human words and machine meaning →
Research areas
Small questions that add up to a safer system shape.
Permissioned tools
How can risky actions stay behind visible approval gates and clear boundaries?
Adaptive routing
How might a system choose verification, fallback, memory, or human review paths?
Internal communication
Can compression-aware representations make internal signals easier to measure and reason about?
Evidence first. Hype last.
COGG9’s research ledger tracks the work honestly: ideas, tests, frozen results, holds, and next steps.
PlanningMemory architecture and visible state boundaries.
TestingStructure behavior, diagnostics, and local hardware constraints.
FrozenResults that are useful, bounded, and not stretched into big claims.
Why local-first matters
AI should be close enough to inspect and bounded enough to trust cautiously.
COGG9 focuses on systems that can run close to the user, respect permission boundaries, and preserve continuity without depending entirely on remote black-box infrastructure.
- Privacy and user control
- Hardware-aware design
- Transparent safety posture
- Lower-cost experimentation
Support
Support page
Support independent AI architecture research.
If this work matters to you — local-first AI, visible safety boundaries, memory-aware systems, and human-guided machine intelligence — there will be ways to follow, support, sponsor, or collaborate as the public materials mature.
Contact / coming soon
Contact page
COGG9 is still being built in public.
Research notes, architecture essays, and support channels will be added as the lab identity and workstreams mature.