README.md

metadata

title: NEXUS OS v4.0
emoji: 🔥
colorFrom: red
colorTo: purple
sdk: gradio
sdk_version: 6.14.0
app_file: app.py
pinned: false
tags:
  - ml-intern

NEXUS OS v4.0 — Intelligent Multi-Provider Router

COMPLETELY self-contained — zero external dependencies except gradio + stdlib. No torch, no pinecone, no package imports that crash on startup.

How It Works

Intelligent Routing (Auto-Detected)

The app queries ALL configured providers in parallel, measures health + latency, and picks the best one automatically. Falls back through the chain if any fail.

Priority	Provider	Free Tier	Strength
1	HF Inference Providers	$0.10/mo credits	Auto-routing, single HF token
2	Groq	Generous	Fastest inference (LPU chips)
3	DeepSeek	5M tokens	Best reasoning models
4	OpenRouter	25+ free models	Most model variety
5	Together AI	Rate-limited 70B	Large models, slow
6	Ollama Relay	Your local models	Via ngrok tunnel
7	Mock	Always works	Simulated for testing

Setup

No setup needed for mock mode. To get real inference, add API keys as Space secrets:

Secret	Provider	Get Key At
HF_TOKEN	HF Inference Providers	Already active in Spaces
GROQ_API_KEY	Groq	https://console.groq.com
DEEPSEEK_API_KEY	DeepSeek	https://platform.deepseek.com
OPENROUTER_API_KEY	OpenRouter	https://openrouter.ai
TOGETHER_API_KEY	Together AI	https://api.together.xyz
OLLAMA_RELAY_URL	Your local Ollama	ngrok http 11434

Features

• 37+ real models in registry
• Thermodynamic telemetry: EEP, PTI, NEWI hallucination signals
• VRAM-aware filtering: only shows models that fit your budget
• Per-token risk scoring: hallucination detection simulation

What's New in v4.0

• Self-contained: no nexus_os_v2/ imports, no torch/pinecone dependencies
• 5 real providers: HF Router, Groq, DeepSeek, OpenRouter, Together AI
• Removed: Kilocode (IDE plugin), OpenCode (IDE plugin), NVIDIA NIM (trial only), Fireworks ($1 credit)
• Intelligent routing: parallel health checks, capability-based model selection

Repository

specimba/nexus-os-v2