Documentation

VISAR Edge Mobile is a zero-trust, local-first command cockpit consisting of two primary endpoints: the native Android Kotlin/Compose deck in your hand and the Python FastAPI companion server on your PC. They establish persistent AES-GCM-256 encrypted WebSocket links and WebRTC video/audio pipelines, completely isolated from third-party clouds.

• Android Handset Deck - Kotlin/Jetpack Compose, hardware execution engine, WebRTC optical feed, OTP ring buffer & holographic trackpad.
• Host Companion Server - Python 3.10+ FastAPI server, system telemetry sampler, command dispatcher, Web Dashboard & MCP stdio/HTTP server.
• AI Agent Clients - Cursor, Claude Desktop, Claude Code, or custom LLM agents utilizing the Model Context Protocol.
• Cyber Web Dashboard - In-browser command center running at http://localhost:8000/dashboard with real-time gauges and terminal runner.

Every frame and REST payload is encrypted end-to-end using AES-GCM-256 with 12-byte IVs, 128-bit authentication tags, and PBKDF2 HMAC-SHA256 derived keys. Legacy plaintext passkeys have been completely eradicated.

# Automated 6-digit PIN pairing flow:
# 1. Host generates short-lived 6-digit PIN with 5-minute expiry
# 2. Handset submits PIN via auto-focus digit slots
# 3. Server issues cryptographically signed session token (Authorization: Bearer <token>)
# 4. Device is registered in trusted_devices.json with granular permissions

The token-bucket rate limiter enforces brute-force protection: 3 consecutive failed authentication attempts immediately triggers a 10-minute network ban.

Built on the Google WebRTC SDK (Android) and Python aiortc backend, Tactical Eye streams real-time desktop video and webcam feeds directly to your phone under DTLS-SRTP encryption with 0% idle CPU overhead.

• PC Display Feed - Real-time screen capture streamed with hardware-accelerated H.264 / VP8 encoding.
• Optical Sensor Feed - Live webcam video streamed to the mobile OLED display.
• Voice Mode - Ultra-low latency acoustic push-to-talk transmission directly into PC audio hardware.
• Hotspot Traversal - Automated ICE candidate rewriting resolves tethered Android hotspot network traversal.

• Ensure Python 3.10+ is installed on your PC host.
• Clone or pull the companion server repository on your host machine.
• Install dependencies: pip install -r requirements.txt
• Launch the FastAPI server and install the signed Android APK.

# 1. Start the desktop companion server
python -m uvicorn mobile_server.main:app --host 0.0.0.0 --port 8000 --reload

# Output:
# [VISAR] Companion Server online: http://0.0.0.0:8000
# [VISAR] Cyber Web Dashboard available: http://localhost:8000/dashboard
# [VISAR] AES-GCM-256 Cryptographic Core active
# [VISAR] MCP Hardware Gateway ready for stdio bridge

Allow port 8000 through your local Windows Defender or Linux firewall for traffic within your home/office subnet.

The companion server exposes an official Model Context Protocol (MCP) stdio and HTTP gateway. Add the configuration to your AI client to immediately grant the agent access to phone tools under visible user control.

{
  "mcpServers": {
    "visar-mobile": {
      "command": "python",
      "args": ["-m", "mobile_server.mcp_stdio"],
      "env": {
        "VISAR_HOST_URL": "http://127.0.0.1:8000"
      }
    }
  }
}

Every hardware tool call is routed through an async Hold & Wake bridge with a 5-second timeout, accompanied by a non-disableable visual banner on the phone screen for transparent security.

Frames between the Android client and the companion server use high-speed newline-delimited JSON (NDJSON) over a persistent AES-GCM-256 encrypted WebSocket link.

// Command execution frame
{ "t": "cmd", "id": "req-9812", "p": { "action": "exec", "target": "open_terminal" } }

// Command acknowledgement frame
{ "t": "ack", "id": "req-9812", "p": { "ok": true, "duration_ms": 18 } }

// Live dual-node telemetry broadcast
{ "t": "telemetry", "p": { "cpu_pct": 14.2, "ram_pct": 68.4, "temp_c": 49.0, "batt_lvl": 94, "current_ma": 420, "charging": true } }

All tools are implemented in Android Kotlin via the AndroidToolEngine singleton with structured error handling and Android 10+ background permission fallbacks.