About this article

• Audience: Integrators connecting Ghost Protocol to vendor SDKs, security frameworks, and cloud video APIs
• Goal: After reading, you know which external specifications and services align with edge encoding, zero-trust federation, and optional cloud escalation
• Type: Reference

Summary

Ghost Protocol builds on industrial edge vision SoCs for on-device inference, OWASP/NIST themes for zero-trust federation, surveillance-style H.265 contracts, and cloud video APIs for justified-clip escalation. Use this page as a curated link map for standards and services — product identity stays Ghost Protocol / Shadow / Phantom Vision.

Prerequisites

• Glossary — precomp, justified recording, Phantom proxy
• Video cloud architecture — end-to-end escalation flow

Edge vision platform (Shadow)

Topic	Ghost Protocol usage
Hardware	Shadow edge cameras (industrial SoC-class vision nodes)
Runtime	Ghost Protocol dual-stream pipelines: RF-DETR DSP, optional depth, tracking
Deploy	Durable edge app bake; limited post-deploy mutation
Models	RF-DETR 640×640, depth layer, streaming patterns in firmware

On-device integration details: Edge AI architecture, Recording.

OWASP / NIST / zero-trust

Theme	Ghost Protocol pattern
No edge secrets	Cloud keys live on Phantom proxy only
Signed federation	HMAC A2A queries; mTLS/HMAC Hermes sidecar
Detection authority	Precomp RF-DETR; graceful offline notes without hallucinated detections
Scenario harness	Mass-transit safety cases for contract verification

Historical cota proof-of-concept federation notes are in appendices/internal-plans/ (internal-only).

ONVIF / video standards (surveillance alignment)

Contract	Requirement
Primary encode	H.265 4K @ ~30 fps with periodic keyframes
Verification	Fps and keyframe alignment scripts in CI and post-deploy probes
Playback	Linear PTS; justified sidecar duration matches wall time
Integration surfaces	Proxy snapshot, pull-recording, playback-ready MP4; MCP tool exposure

See ONVIF reference for proxy-facing event and stream patterns.

Gemini Video + Google Cloud Video Intelligence

Service	When to use	Notes
Gemini File API	Justified ~60 s+ MP4 clips	Upload via proxy pull/transcode; timestamps in prompts; context cache for batch review
Video Intelligence	Structured labels, objects, shot changes	videos:annotate with LABEL_DETECTION, OBJECT_TRACKING; poll long-running ops
Proxy ownership	Always	Proxy holds GCS and API credentials; results fuse to TAP/MQTT/A2A without replacing live precomp

Prompt patterns and rate guidance: Video samples, Operator video guide.

In-repo companion pages

Page	Covers
API quick reference	Proxy HTTP and MCP tool names
MQTT reference	Topic contracts
Model card	RF-DETR precomp contract
RF-DETR ML Pipeline	Training stack, DLC specs, Omni classes, exact S1/S2/VMS model names
Plugin architecture	M8 and third-party plugin boundaries

Next steps

• Plugins and examples — author oak-examples–style extensions
• Checklists — video-understanding gate steps
• Troubleshooting recovery — when cloud escalation fails

Operator depth

Vendor account setup, live API quotas, and integration evidence bundles live in private operator handoff (not published) (private).