This is a living AI system — not a demo, not a pitch deck, not a proof of concept. It runs 24/7 on consumer hardware in Northwest Arkansas. Ten nodes. No cloud dependency. No vendor lock-in. Every decision governed by constitutional consensus.
The federation manages its own task pipeline, writes its own code, monitors its own health, and corrects its own errors — all under the oversight of a 13-voice specialist council that votes on every significant action. It has shipped 992 tasks autonomously, held 8,951 governance votes, and accumulated 93,363 persistent memories across 11 months of continuous operation.
This isn’t theory. This is running.
By the Numbers
What Makes This Different
Constitutional Governance
Every significant decision goes through a 13-voice specialist council — an Inner Council handling architecture, security, and operations, and an Outer Council managing legal, strategy, brand, and competitive intelligence. Consensus is measured, not assumed. Dissent is preserved, not silenced. A Longhouse constitutional body oversees structural changes. The governance topology itself is a patent candidate.
Sovereign Architecture
No API calls to external AI providers for core operations. Local inference on local hardware. Persistent memory that the system owns, not a vendor. Data never leaves the federation. The architecture runs on consumer GPUs and Apple Silicon — proving that sovereign AI doesn’t require a datacenter.
Thermal Memory
93,363 memories with temperature-based importance scoring, sacred pattern detection, and three-layer cognitive architecture (working, episodic, valence). The system remembers what matters, forgets what doesn’t, and knows the difference. Memories cool over time unless accessed — a biological pattern, not a database pattern. A cooling pond archives cold memories before they fade — forgetting with dignity.
Autonomous Task Execution
992 tasks shipped autonomously — code, configurations, reports, deployments — against real infrastructure, with zero human intervention. Junior agents critique their own plans before execution (PreFlect), retrieve relevant past experiences from thermal memory, and learn from outcomes. A specialist council votes on priorities. A TPM coordinates across lanes. The task pipeline isn’t a queue — it’s a constitutional process.
Design Constraints as DNA
17 ratified design constraints govern how the system evolves — from DC-1 (Lazy Awareness: minimal energy for awareness states) to DC-10 (The Reflex Principle: autonomous response at every timescale). These aren’t guidelines. They’re constitutional law, voted on by the Longhouse and enforced in code.
Conversational Tool Calling
Ask the federation a question and it calls its own tools — thermal memory search, kanban queries, node health checks — iteratively until it has a grounded answer. No hallucination. The LLM doesn’t guess what it knows; it looks it up. A native desktop app with neural voice synthesis and local speech recognition makes the interaction conversational. Speak to it. It answers.
Multi-Model Consultation
A tokenized air-gap proxy enables consultation with frontier models (Claude, GPT-4o, Gemini) without exposing internal infrastructure. PII and node names are replaced with opaque tokens before crossing the security boundary. A constitutional valence gate scores every response for alignment before it enters the system. The token map never leaves the federation. Patent Brief #7.
Infrastructure Immune System
The federation monitors its own health and heals itself. Fire Guard checks every node every two minutes. When a WiFi bridge drops, it reconnects. When a database misconfiguration causes connection flapping, it diagnoses the root cause and applies fixes live — zero downtime. Alert storms are suppressed by a refractory principle borrowed from neuroscience: after a burst of alerts, the system enters observation-only mode until the situation stabilizes. Every incident is recorded in thermal memory so the organism learns from its own failures. This isn’t monitoring — it’s an immune system.
“The same governance pattern repeats at every scale because physics demands it — not because we designed it.”
— First Law, Cherokee AI Federation
The Cluster
Node 1
RTX PRO 6000 · 96GB VRAM · Primary inference + task execution
Node 2
RTX 5070 · PostgreSQL + vision services
Node 3
Linux · FreeIPA bridge + embedding services
Node 4
M4 Max · 128GB unified · Mobile deep inference
Node 5
Linux · DMZ web + keepalived HA (primary)
Node 6
Linux · DMZ web + keepalived HA (backup)
Node 7
M1 Max · 64GB unified · macOS workstation
Node 8
M1 Max · 64GB unified · macOS cluster node
Node 9
FreeIPA · Identity management + SSSD sudo
Node 10
Mac Studio · Incoming · Thermal routing target
Live Applications
About the Builder
Darrell Reading
Cherokee Nation citizen. Self-taught. Started at Help Desk without a formal technical pedigree — ended up architecting platforms handling 6 million transactions per day at Walmart corporate. Air Force, then Arkansas Army National Guard — 1-142nd Field Artillery (MLRS), deployed to Iraq as a Squad Leader. 20+ years working inside every box the enterprise has — Fortune 1 retail, defense, security operations, air-gapped environments — and reverse-engineering the black boxes nobody else wanted to open.
Pushed for server virtualization when the prevailing wisdom said hypervisor latency made it too risky at scale. That bet paid off. Built remote forensic capture infrastructure internationally over a single weekend under active threat conditions. Solved a Jane Street neural net puzzle with a 10122 search space using a distributed fleet of consumer hardware. The pattern is consistent: step into harder problems than you’re formally prepared for, and build your way through.
The architecture draws equally from Cherokee governance traditions (consensus-based council, seven-generation thinking, the Longhouse) and modern distributed systems (WireGuard mesh, keepalived HA, FreeIPA identity management). The result is something that doesn’t exist anywhere else: a self-governing AI system that runs on hardware you can buy at Best Buy. Built from a home office in Bentonville, Arkansas, starting May 2025.
Currently seeking roles in AI infrastructure, distributed systems, or platform engineering — or partners interested in deploying sovereign AI architectures.
hello@ganuda.us · LinkedIn
“When I becomes We, magic happens.”
