Home › Architecture
🏗️ Bicameral Architecture
Left hemisphere: madhatter GPU inference. Right hemisphere: Kubernetes cluster. Bridge: Corpus Callosum.
madhatter — Left Hemisphere
| Component | Specification |
|---|---|
| CPU | Intel i7-9700K (8 cores, no HT) |
| RAM | 16 GB — HARD CEILING (R3) |
| GPU | NVIDIA RTX 3060 — 12 GB VRAM. 14b models cannot coexist. |
| OS Drive | Crucial 1 TB NVMe (nvme1n1) → / /boot /home |
| AI Engine Drive | Intel 660p 1.8 TB NVMe → /mnt/ai_engine/ [R2] |
| Tailscale IP | 100.72.190.41 |
| Docker Root | /mnt/ai_engine/docker |
Docker Stack Topology
| Stack | Services | Order |
|---|---|---|
| dryad-core | Caddy (80/443), Portainer (9443), Netdata | 1st |
| dryad-brain | LiteLLM (4000), Redis (6379), Qdrant (6333), open-webui (3010) | 2nd |
| dryad-mcp | mcp-search (8020), mcp-mem0 (8030), mcp-station (8585) | 3rd |
| dryad-apps | OpenWork (3456) | 4th |
| dryad-portal | dryad-portal (3080) | 5th |
| dryad-tutorials | nginx:alpine → this site | 6th |
Systemd services (not in Docker)Ollama (11434) and OpenCode (4096) run as host systemd services. Containers reach Ollama via
host.docker.internal:11434.Inference Request Path
Client → dryad-litellm:4000 (Authorization: Bearer sk-dryad-master)
├── GPU tasks → Ollama host.docker.internal:11434 (RTX 3060)
├── Ollama cloud → qwen3-coder:480b-cloud stub
├── Cloud NIM → NVIDIA NIM API (nim-deepseek alias)
└── Phase 8 CC → cluster 10.10.10.100 (Corpus Callosum)