CLASSIFIED

01 Architecture Overview

DEEPDISH TOP is deployed as a privileged Docker container with full host namespace access — PID, network, and hardware. The container runs a Gunicorn WSGI server with gevent async workers, enabling Server-Sent Events and parallel metric collection across all subsystems.

Container Deployment

Component	Detail
Base Image	python:3.12-slim
WSGI Server	Gunicorn 21.2.0 — gevent async workers, 3 workers, 120s timeout
Network Mode	host — shares host network stack directly
PID Namespace	pid: host — sees all host processes via nsenter -t 1
Privilege	privileged: true — full hardware access (GPU, sensors, Docker socket)
Restart Policy	unless-stopped

Volume Mounts

Host Path	Container Path	Purpose
/var/run/docker.sock	/var/run/docker.sock:ro	Docker API access
/proc	/host/proc:ro	Host process info (CPU, mem, disk)
/sys	/host/sys:ro	Hardware sensor data
/etc/os-release	/etc/os-release:ro	Host OS identification
/run/dbus	/run/dbus:ro	D-Bus socket for systemd
/var/run/avahi-daemon	/var/run/avahi-daemon:ro	mDNS resolution for .local hosts
/mnt/deepdish	/mnt/deepdish:ro	NAS mount (DeepDish storage array)

Python Dependencies

Package	Version	Purpose
flask	3.0.0	Web framework
psutil	5.9.8	System metrics collection
docker	7.0.0	Docker Engine API client
gunicorn	21.2.0	Production WSGI server
gevent	24.2.1	Async worker for SSE/streaming
requests	2.31.0	HTTP client for integrations

02 Backend — app.py (2,017 lines)

mDNS IPv4 Resolution Cache

A custom DNS resolution layer for .local hostnames on the LAN. Required because mDNS can return IPv6 or resolve to the wrong interface on dual-homed machines. Gathers candidate IPs from avahi-resolve, socket.getaddrinfo, and configurable fallbacks. Results are cached with a 120-second TTL and auto-invalidated on connectivity failure.

System Metric Collectors

All collectors read from the host via nsenter -t 1 (PID namespace) or /proc/1/root (mount namespace). Every metric is gathered in parallel using ThreadPoolExecutor.

Collector	Strategy
get_system_info()	Kernel, arch, distro, CPU model, core counts, total RAM — cached indefinitely
get_cpu_info()	Per-core %, total %, frequency, load averages via psutil
get_memory_info()	Total/used/available RAM + swap via psutil
get_disk_info()	3-strategy fallback: /proc/1/mounts → nsenter df → psutil. Filters real FSes only
get_network_info()	Per-NIC I/O, interface details, top 80 active connections
get_processes()	Top 50 processes by CPU% — PID, name, user, RSS/VMS, threads, FDs
get_gpu_info()	NVIDIA nvidia-smi — temps, utilization, VRAM, power, fan speed
get_temperatures()	psutil.sensors_temperatures() — current/high/critical thresholds
get_docker_containers()	All containers with live stats fetched in parallel per running container
get_systemd_services()	Top 50 services via systemctl with D-Bus fallback

Cluster Node Monitoring

Polls Prometheus node_exporter (port 9100) on 6 LAN nodes in parallel via ThreadPoolExecutor(max_workers=6). Metrics parsed from Prometheus text format include CPU, memory, disk, network I/O, load averages, and uptime.

External Integrations

DEEPDISH Intake Portal

Accepts text, URL, or image submissions. URLs auto-fetch <title> and meta description. Images are stored as base64 data URIs. All notes are saved to DEEPDISH/intake/ in the Obsidian vault with YAML frontmatter.

03 API Endpoints

Metric Routes

Route	Description	Concurrency
GET /api/metrics	Fast local-only metrics (CPU, mem, disk, net, procs, docker, systemd, GPU, temps)	12 workers
GET /api/metrics/integrations	Slow external integrations (cluster, NP, urbit, gamer, OCS, HA, obsidian)	8 workers
GET /api/metrics/all	Combined local + integrations (chat context builder)	16 workers

Management Routes

Route	Description
POST /api/docker/<name>/action	Docker control: start, stop, restart, remove
GET /api/docker/<name>/logs	Container log tail (default 80 lines)
POST /api/docker/prune	Remove all stopped containers
POST /api/systemd/<name>/action	Systemd control: start, stop, restart, enable, disable
GET /api/systemd/<name>/logs	journald log tail (default 60 lines)

Obsidian & OCS Routes

Route	Description
GET /api/obsidian/note	Fetch note content (JSON + tags + frontmatter)
GET /api/obsidian/search	Full-text vault search (top 20 results)
POST /api/portal/intake	Submit text/URL/image to vault intake
GET /api/ocs/eye/stream	MJPEG stream proxy
POST /api/ocs/eye/expression	Set eye expression + duration

04 Frontend — Main Dashboard (11,553 lines)

Layout Architecture

Three-column responsive grid with independent scrolling. Left column houses CPU/MEM, network I/O, disk, GPU, temps, storage, interfaces, Native Planet, and system info. Center column contains the cluster view, Obsidian vault, processes, Docker, Radgotchi virtual pet, systemd services, net flow matrix, and connections. Right column holds the OCS eye stream, DEEPDISH chat, topology visualization, Home Assistant, and event log.

Dual-Theme System (Threat Red / Intel Blue)

Two complete color themes toggled by clicking the "DEEPDISH" header text. A radial ripple animation expands from the click origin while all section panels flash with staggered 50ms delays. The class switch is delayed 300ms for dramatic reveal. Theme persists via localStorage.

Visual Effects & Background Layers

• Grid background — Animated CSS grid (40px cells) with slow diagonal drift (20s cycle)
• Wireframe globe — Canvas-rendered rotating sphere (9 lat, 12 lon lines, equator highlight, 2 outer rings). Perspective projection, continuous requestAnimationFrame
• CRT scanlines — Repeating linear gradient overlay (2px lines)
• Kanji watermarks — 神 (kami) and 経 (kei) as fixed-position decorative text

Section Panel System

Every section uses a standardized .section component with corner accent lines, hex pattern overlays, alert pulsing, skeleton loading shimmer, collapse/expand toggles, and GPU-accelerated compositing (translateZ(0), contain: style paint).

Data Visualization

• Sparklines — Inline SVG charts for CPU, memory, network, GPU (60-sample rolling history)
• Progress bars — Color-coded utilization bars (green → orange → red) with animated sweep overlay
• Cluster aggregates — 3 large metric cards (CPU/MEM/DISK) with sweep backgrounds, min/max per metric
• Net flow matrix — Canvas-rendered heatmap of services vs remote hosts. Animated cells, pulse waves, scanline overlay, crosshair on hover
• Obsidian topology — 50-node hex-grid canvas with data packets flowing along connections, comet trails, ambient glow orbs

05 Network Topology Visualization

Force-directed graph rendered on a dedicated canvas element. DEEPDISH sits as the center-pinned hub node with cluster nodes on an equidistant inner ring and external integrations on an outer ring. Full physics simulation with spring forces, repulsion, velocity damping, and boundary clamping.

Node Type	Name	Layout
host	DEEPDISH	Center-pinned hub (diamond SVG)
cluster	mini, cheese, pepper, onion, garlic, supreme	Equidistant inner ring (Apple logo)
tellurian	Native Planet	Outer ring
urbit	Cloud Urbit	Outer ring
pc	Epic Gamer	Outer ring (Windows logo)
homeassistant	Home Assistant	Outer ring (HA gear icon)

06 Radgotchi Virtual Pet System

A complete virtual pet IIFE module (RG) displayed in the RADTECH section. 25 sprite states, 12 CSS animations, military/SIGINT-themed status messages (e.g., "SIGINT NOMINAL", "OVERWATCH ACTIVE", "EXCESSIVE INPUT — CEASE").

07 OCS Eye & DEEPDISH Chat

MJPEG Eye Stream

OCS camera feed from chef.local:5001 proxied through Flask to avoid CORS. Health check every 10 seconds, auto-reconnect on error with 5-second delay. 7 expression controls (happy, sad, angry, bored, confused, surprised, suspicious). System events automatically trigger expressions with cooldown.

AI Chat Integration

Setting	Value
LLM Endpoint	supreme.local:1234 (LM Studio)
Model	gemma-2-27b
Temperature	0.7
Max Tokens	300
Persona	DEEPDISH — black-box intelligence, SIGINT style, lowercase/clinical

The system prompt dynamically injects full live telemetry context — system info, CPU/memory/GPU/disk/network, Docker containers, systemd services, cluster nodes, Home Assistant, OCS, and Obsidian vault stats. Personal names are redacted (OPERATOR / ASSET-2) in chat history.

08 Event Detection & Logging

The event system compares current vs previous data snapshots to detect state changes. Events are classified into 4 levels and trigger both Radgotchi reactions and OCS eye expression changes:

Condition	Level
CPU >90%	● CRITICAL
CPU >70%	● WARNING
Memory >90%	● CRITICAL
Disk >90%	● WARNING
Temperature > critical threshold	● CRITICAL
Docker container state change	● INFO
Cluster node offline	● WARNING
Systemd service failure	● WARNING
Service recovery	● OK

09 Polling & Refresh Architecture

Split polling strategy separates fast local metrics from slow external integrations:

Interval	Target	Purpose
2,000ms	/api/metrics + /api/metrics/integrations	Core telemetry polling (parallel)
1,000ms	—	Column height sync
3,000ms	Management logs	Docker/Systemd auto-refresh
6,000ms	—	Radgotchi idle behavior check
10,000ms	/api/ocs/eye/ping	OCS stream health check
rAF	Canvas elements	Globe, topology, net flow, Obsidian topo

10 Dashboard Preview

11 Performance Optimizations

Optimization	Detail
Split API polling	Local metrics (fast) and integrations (slow) fetched in separate parallel requests
ThreadPoolExecutor	Container stats, cluster nodes, OCS status — all fetched in parallel
DOM cache	Map-based ID lookup cache avoids repeated getElementById calls
CSS containment	.section { contain: style paint } — limits layout recalculation
GPU compositing	translateZ(0) + backface-visibility: hidden on animated elements
Static info cache	System info (kernel, CPU model) cached indefinitely after first fetch
Obsidian TTL cache	30-second cache prevents API hammering on every poll cycle
mDNS cache	120-second TTL with connectivity validation
Connection pooling	Shared requests.Session with keep-alive for all external HTTP
Conditional render	Row flash animations only fire on actual value changes

12 Security Posture

13 External Service Dependencies

Service	Host	Protocol	Purpose
Prometheus (×6)	{node}.local:9100	HTTP/Prometheus	Cluster node metrics
Native Planet	nativeplanet.local:9100	HTTP/Prometheus	Urbit machine metrics
Home Assistant	homeassistant.local:8123	HTTP REST	Smart home / phone tracking
Obsidian REST API	localhost:27124	HTTPS REST	Vault access
OCS	chef.local:5001	HTTP + MJPEG	Eye rendering / vision
LM Studio	supreme.local:1234	OpenAI API	AI chat (gemma-2-27b)
Docker Engine	/var/run/docker.sock	Unix socket	Container management
systemd	D-Bus	D-Bus / nsenter	Service management