abonaie222263/model-hub-chat

Model Hub Chat 💬

A professional, macOS-style chat app where you can talk to any model — either running entirely in your browser (free CPU, private, no account) or through any OpenAI-compatible endpoint (Hugging Face Inference Providers, opencode, Hermes, …).

Features

• —🧠 In-browser inference — 8 curated models run locally via transformers.js (ONNX/WASM). Zero servers, zero cost, 100% private.
• —🔌 Remote API mode — stream tokens from any /v1/chat/completions endpoint, with curated HF-hosted presets (openai/gpt-oss-20b, Qwen/Qwen2.5-72B-Instruct, DeepSeek-V3, …) and custom base URLs.
• —⚙️ Full settings page — completion endpoint (base URL + API key + model), system prompt, temperature / max tokens / top-p, plus quick sliders right above the composer.
• —🔑 Bring-your-own token — paste an HF token (or pass ?hf_token=…) to use its credits on hosted models.
• —🤖 opencode / Hermes integration — the Settings page generates a ready-to-paste opencode.json provider config plus the raw endpoint URL.
• —🖥 macOS-style UI — traffic-light title bar, sidebar, clean typography.

Models

In-browser (free, private)

Remote API (HF Inference Providers, your token)

• —openai/gpt-oss-20b (default)
• —Qwen/Qwen2.5-72B-Instruct
• —meta-llama/Llama-3.3-70B-Instruct
• —deepseek-ai/DeepSeek-V3
• —mistralai/Mistral-Nemo-Instruct-2407
• —HuggingFaceH4/zephyr-7b-beta (free tier)

Why this approach?

A Static Space was chosen because free accounts can't host a server-side Gradio/Docker Space (needs PRO), and the visitor-pays OAuth pattern needs an OAuth app. This Space works for everyone immediately:

• —No server, no quota — in-browser inference uses the visitor's own CPU.
• —Optional API — connect any OpenAI-compatible endpoint for bigger models.

Usage

1. 1.Open the app.
2. 2.Pick a model (Models tab) or switch to API mode.
3. 3.Tune the system prompt / endpoint / generation in Settings.
4. 4.Chat.

Connect opencode / Hermes

In Settings, copy the generated endpoint URL or the opencode.json config. Example:

{
  "provider": "modelhub",
  "base_url": "https://router.huggingface.co/v1/chat/completions",
  "api_key": "<YOUR_HF_TOKEN>",
  "model": "openai/gpt-oss-20b",
  "chat": { "temperature": 0.7, "max_tokens": 512, "top_p": 0.95 }
}

License / credits

UI and code: MIT. Models keep their own licenses (Apache 2.0, MIT, Llama 3.2, Gemma). Built with transformers.js.