X (Twitter) | Contract: [TBD]
Whispr is a high-performance, real-time AI whispering system designed as a cognitive companion for professionals. Operating through a terminal-inspired interface, Whispr serves as a silent strategic coach that captures live audio streams and provides instantaneous, context-aware "whispers" tailored to the user’s specific identity.
Whispr utilizes a sophisticated multi-model pipeline to ensure sub-second response times and high-fidelity reasoning.
graph TD
A[Live Audio Stream] --> B{Audio Capture}
B -->|System Audio| C[Deepgram Nova-2 / Groq Whisper]
C -->|Real-time STT| D[Transcription Buffer]
D -->|Silence Detection 2s| E[AI Orchestrator]
F[Shadow Profile JSON] --> E
E --> G{Model Router}
G -->|High Speed| H[Groq LPU™ Llama 3.3]
G -->|Deep Reasoning| I[Google Gemini 2.0 Flash]
G -->|Specialized| J[OpenRouter / Qwen]
H & I & J --> K[Contextual Whisper Synthesis]
K --> L[Terminal UI / Stealth Overlay]
sequenceDiagram
participant U as User
participant S as System Audio
participant STT as Transcription Engine
participant C as Composables (useWhispr)
participant AI as AI Model (Groq/Gemini)
participant UI as Terminal UI
U->>S: Starts Capture
S-->>STT: Pulse PCM Stream
STT-->>C: Real-time Transcript
C->>C: Monitor Silence (2s)
Note over C: Threshold Reached
C->>AI: Send Context + Shadow Profile
AI-->>C: Generate Tactical Insight
C->>UI: Render "Whisper" Card
UI-->>U: Silent Alert
Utilizing Deepgram’s Nova-2 and Groq Whisper, Whispr achieves sub-second audio-to-text conversion. The system processes streams in real-time, feeding a rolling buffer that monitors for conversational gaps.
The AI doesn't just react; it waits for the perfect moment. Upon detecting 2 seconds of silence, the system triggers a synthesis cycle. It cross-references the latest 10-15 messages with the user’s Shadow Profile to generate a response that is "one-take ready."
The core of Whispr's personalization is the Shadow Profile—a JSON-defined identity that guides the AI's tone, knowledge base, and strategic goals.
{
"identity": {
"nickname": "Alex",
"title": "Senior Engineer",
"background": "Focus on high-performance distributed systems..."
},
"technical_skills": {
"languages": ["TypeScript", "Go", "Rust"],
"expertise": ["System Design", "Real-time Data"]
}
}├── components/
│ ├── landing/ # Marketing components
│ ├── shared/ # Common UI (CyberBackground, GlitchCursor)
│ ├── ui/ # Base UI elements (Modals, Confirmations)
│ └── workspace/ # Core App logic (Terminal, AI Panel, Video)
├── composables/
│ ├── useAIWhisperer.ts # AI Orchestration logic
│ ├── useTranscription.ts # STT Bridge (Deepgram/Groq)
│ └── useWhispr.ts # Global State & Persistence
├── assets/
│ └── personality.json # Default Shadow Profile
├── pages/
│ ├── index.vue # Landing Page
│ └── workspace.vue # Main Application View
└── public/ # Static assets & icons
- Frontend: Nuxt 3 (Vue.js) - Optimized for SSR and fast hydration.
- Styling: Vanilla CSS with TailwindCSS for utility-first layout management.
- Real-time Comms: WebSockets / MediaDevices API for high-fidelity audio capture.
- AI Infrastructure:
- Groq LPU™: Optimized for extreme speed.
- Gemini 2.0: High-context reasoning.
- OpenRouter: Access to a diverse model ecosystem.
- Persistence: Zero-Server Architecture using LocalStorage for keys and history.
- Node.js (v18.x or later)
- API Keys: Deepgram, Groq, Google Gemini, OpenRouter
-
Clone & Install:
git clone https://github.com/cuda-cookie/whispr.git cd whispr npm install -
Environment Configuration:
cp .env.example .env # Add your API keys to .env -
Launch:
npm run dev
- Zero-Server Persistence: All conversation history and Neural Profiles are stored locally in the browser.
- Local Key Management: API keys are processed client-side and never stored on external databases.
- Protocol Purge: Instant memory wipe command available via terminal interface.
Built with technical excellence by Whispr Assist © 2026
Follow us: X (Twitter) | GitHub