🔥 Repo Roaster

Paste any GitHub URL. The framework fetches the repo data, indexes it into the local Orama vector DB, and streams a Senior Architect roast — showcasing Local RAG + Speculative Streaming.

Local RAG Speculative Stream

🔗 GitHub Repository URL

Draft Stream Speculative Stream

🔥 Waiting for a GitHub URL…

🔍 RAG Context Retrieved

📚 Vector search results will appear here

🧠 Knowledge Nexus

Index text or URLs into the live Orama vector database and watch chunks appear as nodes on the 2-D local retrieval map. Query to see similarity lines.

Vector DB Lite Embeddings

✍️ Index Text

🌐 Index URL

🔍 Hybrid Query

📚 Indexed Sources

📭 No documents indexed yet

Nodes: 0

Last embed: —

💬 Sovereign Chat

Chat with a local-only LLM. Toggle "Technical Overlay" to reveal which tokens came from the Draft model vs the Speculative target.

100% Local Token Transparency

🔒 Local Inference — Zero Data Egress

Draft: 0 Spec: 0 Local: 100%

⚡ Multi-Tab Stress Test

Launch 5 browser windows, each running simultaneous inference. The VRAM meter stays flat — proving all tabs share one SharedWorker instance.

VRAM Dedup SharedWorker

🖥️

Global VRAM Deduplication

5 windows. 1 ONNX model instance. The Multiplexer routes all requests through a single SharedWorker — the model never loads twice.

⬜Tab 1

⬜Tab 2

⬜Tab 3

⬜Tab 4

⬜Tab 5

📊 Live Metrics 0

📋 Verification Artifact

🔬 Run the stress test to generate a verification report