Sovereign Intelligence
Each node runs its own AI, accumulates sensor data, and operates indefinitely with zero connectivity. No cloud. No permission. No phone-home. The node doesn't report to a network — it is a complete nervous system.
FrogNet · BLDC-1 · AI Hosts
FrogNet is a community of autonomous intelligences that think together when they can and think alone when they must. Each node is a complete, sovereign computer — its own TCP/IP network, DNS, web server, database, and AI. When nodes meet, they share awareness. Like experts arriving at the same crisis, new intelligence emerges that didn't exist anywhere a moment before.
Why this exists
Your heart has its own pacemaker cells. Your gut has its own nervous system. Your spinal cord handles reflexes before your brain even knows something happened. The brain coordinates and provides higher-order intelligence — but it does not run the body. That's FrogNet. Each node senses its environment, processes what it finds, acts on what it knows — locally, immediately, without asking anyone. The sensor cluster is the nerve ending. The local AI is the spinal reflex. When nodes connect, you don't get a dumb pipe between a sensor and a distant brain. You get two nervous systems sharing awareness.
Each node runs its own AI, accumulates sensor data, and operates indefinitely with zero connectivity. No cloud. No permission. No phone-home. The node doesn't report to a network — it is a complete nervous system.
When nodes discover each other — over radio, WiFi, Ethernet, WireGuard tunnels, or any combination — they automatically mesh. Trigger rules govern what crosses the link: critical alerts immediately, bulk history when bandwidth allows.
The transient database is intentionally ephemeral — a shared scratchpad, not a source of truth. When the network fragments, each piece keeps working. When fragments rejoin, there's no reconciliation pause. The database matches the physics.
BLDC-1 Semantic Compression
BullFrog Long Distance Communications Protocol strips all HTTP framing — no method line, no headers, no status codes. The system learns the shape of actual traffic empirically, builds templates automatically, and caches the most recent response.
Nothing changed. The response is identical to cache. One MeshCore packet with room to spare. A 10KB sensor response becomes 20 bytes.
Some fields changed. A bitmap identifying which fields changed, followed by only the changed values. A 37-field sensor reading where 10 things changed: ~275 bytes after LZW.
Everything changed or new endpoint. Full payload, but still stripped of all HTTP framing and LZW-compressed. Still dramatically smaller than raw HTTP.
6× effective throughput at first contact; approaching 15× in steady learning states with 20% packet loss. Five Raspberry Pi nodes (BlackBox, IronBox, SilverBox, HardBox, TealBox) forming a self-healing mesh over Ethernet, WiFi, and WireGuard tunnels. Every node runs its own DHCP, DNS, web server, MySQL database, and local AI.
The architecture metaphor
When two FrogNets come within range, it's not a client finding a server, or a terminal reaching a mainframe. It's two experts walking into the same room. Each one already has knowledge, context, judgment, and history. Neither needs the other to function. But when they meet, something new becomes possible.
One has been watching river levels. The other has been monitoring structural vibration on a bridge. Each has been running its own AI, drawing its own conclusions. When they mesh, the bridge node suddenly has upstream water data. Its AI correlates rising water with vibration anomalies it noticed but couldn't explain. The river node now knows there's a vulnerable bridge downstream. New intelligence emerges that existed nowhere before.
The meteorologist who knows another storm is coming. The picture changes again. This is the Living Network. Not a system that transmits data. A community of intelligences that think together when they can and think alone when they must. It grows when you add nodes, shrinks when you lose them, splits when links fail, heals when they come back. Every state between fully connected and fully isolated is functional.
Where FrogNet works
FrogNet was born as DisasterComm — disaster-area mesh infrastructure. It evolved into a general-purpose Private Internet for any environment where connectivity is unreliable, bandwidth is constrained, privacy is non-negotiable, and operational continuity matters more than convenience.
GPS tracking, chat, photos, presence — all on hardware you own. No cloud. No ads. No data harvesting.
Full TCP/IP web applications over 4800-baud RF links. Dashboards, databases, sensor aggregation — the tools an EOC actually needs.
Activity heuristics and anomaly detection running locally, no camera surveillance. AI that recognizes patterns and alerts caregivers.
ESP32 soil, weather, and plant sensors aggregated by Pi Zero concentrators. AI-driven decisions made locally.
Sovereign mesh for environments where cloud dependency is a vulnerability. Controlled propagation, fail-closed posture.
SCADA, substations, remote infrastructure. Sensor-to-AI-to-actuator loops that run offline across radio links.
Running today
BlackBox, IronBox, SilverBox, HardBox, TealBox — five nodes on Raspberry Pi hardware forming a complete mesh. Every node runs its own DHCP, DNS, web server, MySQL database, and local AI.
WireGuard tunnel lifecycle — creation, refresh, reconnect, teardown. Nodes behind NAT reconnect and rebuild transparently. WebSocket connections from field nodes. Mesh topology orchestration.
Native Android (Kotlin) with GPS beacon broadcasting, real-time map, text chat, WebRTC voice/video, sticky notes, and a custom OkHttp DNS resolver for .frognet hostnames.
Ed25519-signed boot-time script distribution with TOFU trust, nonce replay protection, and gossip relay. Push from one node; every node executes at next boot, cryptographically verified.