TTS cache
Repeated phrases — greetings, sector intros, recurring closers — play back from disk. Zero-latency, zero-cost.
Voice Agent
Cost-aware · Cached · Arbitraged
Speak with your market intelligence
A real-time conversational agent for Indian equity sectors. Engineered to be fast, to be premium, and most of all — to never leak money on silence, repeats, or the wrong region.
Voice Agent
Ready to Connect
Conversational Market Intelligence
Tap the 3D Aura Orb or press Start to initiate live bidirectional voice streaming.
Warming up the cache…
Tap orb to connect
Acoustic Harmonic StreamMicrophone Input
Suggested Market Inquiries
Tap the orb to start a voice conversation
Ask about a sector, a stock, or a market move. The agent listens, reasons, and speaks back — with cached responses costing nothing on repeat.
Warming up the cache…
Voice Intelligence Capabilities
- •Deepgram Agent + HF Speech-to-Speech: Ultra-low latency turn-taking, barge-in detection, and real-time streaming speech synthesis.
- •Tool Execution Engine: Live pricing for Indian stocks (NSE/BSE), crypto, portfolio balance lookups, and mock trade orders.
- •Hinglish Fluency: Natural understanding of Hindi-English financial slang (bhaav, kharidna, bechna, girawat).
- •Cost-Aware Caching: Common questions and briefings hit local disk audio cache at zero API cost.
Five places where voice agents leak money — and how we plug them
Voice activity detection
Silence, breath, and dead air get trimmed before transcription so we never pay to think about silence.
Tight responses
Voice replies are capped server-side and trimmed at sentence boundaries — short answers, lower bills.
Prompt caching
The system prompt is static, so upstream providers cache the prefix and only bill us for the new turn.
Regional arbitrage
Latency is sampled per provider every turn. The fastest healthy one wins; degraded ones go on cooldown.