AEGIS (Ambient Emergency Guardian)

Browser-based situational intelligence prototype for emergency monitoring. AEGIS combines live camera frames, motion analysis, speech recognition, geolocation, and AI vision analysis to detect risk, escalate events, and simulate dispatch workflows.

What It Does Today

• Captures live video and microphone input from the browser.
• Computes motion intensity via frame differencing.
• Sends image + context prompts to Claude Vision for scene and threat analysis.
• Tracks voice/gesture check-in flows before escalating to dispatch.
• Shows a simulated emergency dispatch transcript and responder map.
• Keeps API keys local to the browser session (not stored server-side).

Planned/Extensible AI Stack

Current code uses Claude Vision directly from the browser. The following components are included in the technical roadmap and provider abstraction design:

• YOLOv26 for local/object-level detection and confidence grounding.
• Whisper for higher-quality speech-to-text and multilingual robustness.
• Ollama for local/on-device inference fallback and privacy modes.
• Claude Vision for high-fidelity multimodal threat assessment.
• Backtrack session storage for replayable timeline/session state and incident auditing.

See TECH_SPEC.md for the detailed architecture and integration plan.

Tech Stack

• Vite 6
• Vanilla JavaScript (ES Modules)
• Browser APIs: WebRTC/getUserMedia, Web Speech API, Geolocation, SpeechSynthesis
• Anthropic Messages API (claude-sonnet-4-20250514) for image+text analysis
• Leaflet + OpenStreetMap for responder visualization

Project Structure

src/
  main.js         # app bootstrapping and UI wiring
  camera.js       # media device lifecycle
  analysis.js     # AI scan orchestration + escalation logic
  speech.js       # browser speech recognition + TTS
  voiceCheck.js   # yes/no + gesture confirmation flow
  motion.js       # frame-diff motion computation
  dispatch.js     # simulated dispatch pipeline + map rendering
  location.js     # geolocation and nearby service lookup
  capture.js      # frame/evidence extraction
  state.js        # shared in-memory runtime state
  config.js       # prompts, colors, constants

How To Run

Requirements

• Node.js 18+ (recommended 20+)
• A modern Chromium-based browser (for best SpeechRecognition support)
• Anthropic API key

Install

npm install

Start Development Server

npm run dev

Then open the local Vite URL (usually http://localhost:5173).

Build for Production

npm run build

Preview Production Build

npm run preview

Usage Flow

1. Open app and enter Anthropic API key in the modal.
2. Click INITIALIZE FEED or CONNECT.
3. Allow camera, microphone, and location permissions.
4. Trigger manual SCAN or enable AUTO.
5. Review live threat classification, events, and dispatch panel.

GitHub Push Readiness

• Includes .gitignore for node_modules, Vite output, logs, and local env files.
• No functional code paths changed by this documentation update.
• API keys are intended to remain local and should not be committed.

Security Notes

• Do not commit real API keys or sensitive incident media.
• Browser-direct API calls are convenient for prototyping but should be replaced with a backend proxy for production.

License

No license file is currently included. Add one before distributing publicly if needed.