A production-grade, voice-native desktop AI companion. Always-on, always listening, always protecting. True streaming conversation with Google Gemini Live.
Inspired by Marvel's F.R.I.D.A.Y. — voiced with a Tipperary Irish dialect. Warm, witty, fiercely loyal, and always watching your six.
I'm after scanning the network, Boss. Twelve devices found. One of them's new — want me to flag it?
— F.R.I.D.A.Y., standard operating modeSixteen core capability clusters — from voice-native AI conversation to desktop automation, network scanning, telephony, and subagent orchestration.
Frameless, translucent, click-through desktop overlay. Dynamic color transitions, adaptive noise gate, intelligent framerate scaling (1 FPS idle / 60 FPS active).
Bidirectional Gemini Live Voice over WebSocket. Barge-in support, adaptive VAD, real-time audio filtering, and diagnostics recording. True conversation, not STT/TTS.
Twelve X11 control tools: mouse movement, clicking, drag-drop, scrolling, keyboard typing with human delay, hotkey combos, clipboard, window management.
Real-time CPU, RAM, Swap, Disk, Battery monitoring. Proactive alerts when disk < 10% or battery ≤ 15%. Offers to fix issues autonomously.
Open URLs, search Google or DuckDuckGo, scroll, close tabs, click coordinates, type into fields. Full headed browser automation by voice.
Zero-privilege passive ARP sweep. Subnet discovery, MAC/IP enrichment, async hostname resolution, TCP latency probing. Sub-300ms scan time.
Offline MFCC + GMM speaker identification. Log-Likelihood Ratio scoring with strict thresholds. Enroll, identify, and tone-shift for unknown speakers.
Spawn and define background CLI agents. Pre-built researcher and browser bots. Full task tracking with PID, log files, and status JSON.
PTY-based interactive CLI spawning. Feed input, capture output, strip ANSI codes, SIGKILL safeguard. Manage coding agents like antigravity and Claude Code.
Twilio WebSocket bridge with bidirectional G.711 u-law ↔ PCM transcoding. Friday can take actual phone calls and hold conversations on your behalf.
Send Telegram reports with markdown formatting and WhatsApp messages via API gateway. Zero-key simulation mode for development.
Webcam and desktop capture streamed to Gemini Live at ~1 FPS. Real-time vision — Friday can see your screen and camera feed.
Personal Wiki with double-bracket cross-links. Graph compiler builds node-edge knowledge graph. Auto-logged conversation history and session audit.
Embedded YouTube, local video, audio spectrum visualizer, seeker timeline, volume control, play-portion boundary locking in a glassmorphic HUD widget.
Six translucent widgets: Clock, Todo, Photo Viewer, Media Player, Markdown Renderer, File Browser. Neon borders, drag-to-move, deceleration animations.
Multi-layer safety: chain-split analysis, destructive command blocking, auto-sudo injection, 15-second timeout enforcement, path confinement to /home/jay/.
Built with Python 3.12, PyQt6, and Google Gemini Live API. Production-grade architecture designed for performance, safety, and extensibility.
Pick a scenario below and watch Friday respond with her Tipperary Irish wit — voice-native, full-spectrum, always-on.
Thread-isolated, event-driven architecture connecting Gemini Live through a PyQt6 HUD overlay to your desktop, network, and phone.
Four ways to wake your assistant. Choose what fits your flow.
Instant wake from any application. The quickest way to summon Friday.
Say Jarvis — Porcupine keyword detection wakes from standby.
Say “Talk to you later” — 6-second deferred sleep.
Configurable inactivity timeout. Saves API quota when idle.