📊 Deep-Chat Telemetry

┌────────────────────────────────────────────────────────────────────────┐
│                        📊 TELEMETRY COMPRESSION BOARD                  │
├───────────────────┬──────────────────────┬─────────────────────────────┤
│ 📦 FOOTPRINT       │ 🚀 PERFORMANCE       │ 💥 STABILITY & ERRORS       │
│ • Lit/Vanilla Web │ • Client-side LLMs   │ • Direct API Failures       │
│ • Multi-Framework │ • Realtime Audio STT │ • Key Exposure Risk         │
└───────────────────┴──────────────────────┴─────────────────────────────┘

---

📈 PERFORMANCE

• OpenAI Realtime Latency: Direct WebSocket -> Sub-100ms Speech-to-Speech — Connected via Native Stream integration.
• Client Inference Boot: Remote API -> Local WASM Web LLM — In-browser hosting via deep-chat-web-llm.
• Stream Hydration: Chunked Response -> HTML Wrappers — Allows progressive UI rendering for live tokens.

---

📦 FOOTPRINT

• Core Package: deep-chat (Web Component / Custom Element) — Bundled via Rollup + Vite.
• Framework Adapters: React wrapper package decoupled to react.
• Server Overhead: 0 KB — Local prototyping via directConnection properties.

---

💥 STABILITY

• API Request Resiliency: Interceptor pipeline captures and reformats incompatible payloads before transmission.
• Webcam/Mic Permissions: Error fallback handles denied media streams gracefully.

---

🚨 GOTCHAS

• Direct Connections Exposure: Setting directConnection exposes API keys in browser network traffic. Use ONLY for local prototyping. Switch to backend proxies (Node, Python, Go templates in example-servers) for production.
• Web LLM Assets: Browser-hosted models require loading large external model weights, which impacts initial page bandwidth.