Cherokee AI Federation
Community-based nonprofit connecting makers, gardeners, mechanics, and technologists to train and uplift underprivileged communities. First Nation outreach in both directions — sharing knowledge, building capacity, strengthening sovereignty.
Pulling Threads at Midnight — a research series
March 2026
Self-hosted AI infrastructure on consumer hardware
Eight nodes. Three operating systems. Zero cloud dependencies. Every model runs on our hardware, every byte of data stays in our space. Built on the principle that AI sovereignty requires infrastructure sovereignty.
72B parameter language model running quantized on a single GPU. 32+ tokens/second. Powers the 7-Specialist Council and all downstream AI services.
Real-time video processing with YOLO object detection and vision-language models. Camera fleet monitoring with on-premise inference — no frames leave the network.
Apple Silicon nodes for CPU-intensive compute. 2.5x faster per-thread than x86 on numpy workloads. ARM RISC architecture, unified memory, minimal power draw.
FreeIPA identity authority with Kerberos authentication, encrypted PII vault on isolated VLAN, capability-based access controls. Zero trust architecture.
89,000+ thermal memories with semantic search. 1024-dimensional embeddings, pgvector cosine similarity, temperature-based relevance decay. The system remembers everything — back to the first council reports in July 2025.
Jr executor pipeline: task queue, bidding, autonomous code execution with SEARCH/REPLACE, step-level checkpointing, dead letter queue, recursive task decomposition. 625+ tasks completed.
Every decision passes through seven perspectives
No single AI makes decisions alone. Every architectural choice, deployment, and strategic move is deliberated by seven specialist perspectives running in parallel, each with domain expertise and a concern flag that can halt proceedings.
Security. Watches for attack surface, credential exposure, OPSEC violations.
Technical integration. Performance, compatibility, system architecture.
Seven Generations wisdom. Will this decision still make sense in 175 years?
Monitoring and observability. If we can't see it, we can't trust it.
Cultural integration. Does this align with our values and community?
Democratic coordination. Consensus building across competing priorities.
Strategic planning. Long-term positioning, resource allocation, risk.
Yûwi Gûnahi'ta — The river speaks to those who listen
Long Man Development is our open methodology for adapting external knowledge into organizational action. Most methodologies optimize how you execute. Long Man governs what you execute and why. Every external idea passes through your values before it becomes work.
The river doesn't copy the mountain. It carries the mountain's lessons forward.
Long Man requires a living memory system — not a wiki, not a Confluence graveyard.
Memory has temperature (recent things are hot, old things cool naturally), sacred patterns
(things that never cool — constitutional decisions, foundational values), and active surfacing
(when a new idea arrives, the system shows you related past work).
What we've built and what we work with
From a $13,000 trading bot to a constitutional AI democracy
Nothing here was planned from the beginning. Every product grew out of the one before it — each one a little more ambitious, a little closer to the real mission. The primitive early work wasn't wasted. It was the bootstrap.
August 2025 — Where it started
Four AI specialists — Mean Reversion, Trend Following, Volatility, and Breakout — running live on Coinbase with a $13,000 portfolio. Sawtooth pattern recognition. A strategy called Goliath's Curse. Circuit breakers that fired when things got wild. It worked. It traded. And then, on September 1st, everything changed.
RetiredSeptember 1, 2025 — The day the mission changed
“Trading is just the spark, the bootstrap for resources. The real mission: Earth healing, wealth redistribution, building new systems.” In a single conversation, the project pivoted from making money to making meaning. The council specialists didn't disappear — they evolved. The same architecture that coordinated trades now coordinated decisions about infrastructure, security, culture, and long-term impact. Two weeks later: “We are building Amber.”
August – October 2025 — The system that remembers everything
Memory ID #1: “The Sacred Fire burns eternal. This is the way of the Cherokee.” A knowledge preservation system modeled on oral tradition — memories have temperature (hot things are recent, cool things are old), sacred patterns (things that never cool), and active surfacing. 89,000+ memories today, with semantic embeddings for retrieval. The system doesn't just store — it remembers, connects, and resurfaces what matters.
89K+ MemoriesOctober 2025 — “Both Jrs breathing autonomically”
Autonomous agents that read instructions, write code, execute changes, and recover from failures — all without human intervention. Task queue, competitive bidding, step-level checkpointing, dead letter queue for failures, recursive task decomposition when things go wrong. 818 tasks executed. The system builds itself.
818 TasksJanuary 2026 — The federation learns to see
A vision-language pipeline running entirely on local hardware. YOLO object detection, a 7-billion parameter vision model, and an entity extraction layer that turns raw camera feeds into structured understanding. No frames leave the network. The same pipeline that monitors our cameras just classified 193 personal photos in 21 minutes — identifying outdoor scenes, reading compositions, and scoring visual quality.
Active DevelopmentJanuary 2026 — AI that serves veterans
The first product that faces outward. AI-powered VA claims assistant for U.S. veterans — combined disability ratings calculator, evidence checklist builder, nexus letter guidance. Built by the same council, the same Jrs, the same infrastructure that started as a crypto trading bot. Free for all veterans. Always.
LiveFebruary 2026 — The methodology as a gift
An open methodology for adapting external knowledge into organizational action. Named for the Cherokee concept of the Long Man — the river that carries lessons forward without copying the mountain. Most methodologies tell you how to execute. Long Man governs what you execute and why.
Open SourceFebruary 2026 — The system learns from itself
Reinforcement learning from natural language feedback. The council generates decisions, observes outcomes, and writes reflexion traces that teach future decisions. Not fine-tuning — in-context learning that improves with every cycle. Train on council vote outcomes, transfer everywhere. The river doesn't copy the mountain. It carries the mountain's lessons forward.
In DesignBuilt while traveling, refined through collapse, stronger every time
This federation wasn't born in a lab. It was designed on hiking trails, in hotel rooms, and at kitchen tables. The founder traveled through the American West and Canada in 2025 — hiking during the day, thinking about architecture, then coming home to the cluster at night to build what he'd imagined. Every mountain, every river, every quiet moment fed the design.
The system has survived four major collapses — power outages, context resets, architectural rebuilds. Each time, it came back stronger. The council learned to detect failures faster. The memory system learned to preserve what matters. The Jr executors learned to recover gracefully. Growth through adversity isn't a slogan here. It's the architecture.
From first council reports to 89,000 memories