Upload README.md

README.md

@@ -1,10 +1,28 @@
 ---
-title: Muse Glimmer 30b Agentic Demo
-emoji: 📊
+title: Muse Glimmer 30B Agentic Demo
+emoji: ✨
 colorFrom: purple
-colorTo: red
+colorTo: indigo
 sdk: static
-pinned: false
+app_file: index.html
+short_description: Safe tool-call demo for Muse Glimmer 30B
+models:
+  - vcruz305/Muse-Glimmer-30B-Hermes-Agentic
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# Muse Glimmer 30B Hermes Agentic demo
+
+This Space demonstrates [`vcruz305/Muse-Glimmer-30B-Hermes-Agentic`](https://huggingface.co/vcruz305/Muse-Glimmer-30B-Hermes-Agentic). The live static interface can call a public, no-auth OpenAI-compatible endpoint serving the model. A self-hosted ZeroGPU Gradio implementation is included in `app.py` and is ready to activate when GPU hosting is available.
+
+The interface demonstrates the behavior targeted by the fine-tune: answer directly when a tool is unnecessary, or propose one or two Hermes-style tool calls and stop. Tool calls are parsed and displayed, but never executed.
+
+## Safety and limitations
+
+- The Space does not run commands, read files, browse the web, or perform destructive actions.
+- Outputs are generated by research weights and may be incorrect or unsafe.
+- The model is about 30B parameters and requires a 96 GB ZeroGPU allocation in BF16. The account that owns this Space is not currently eligible to create ZeroGPU Spaces, so the static client does not host the weights itself.
+- The model card reports 16/20 on its simulated loop gate and 7/7 on a small native Hermes battery. These results are not a product certification.
+
+## API and MCP
+
+The static demo accepts a public, no-auth OpenAI-compatible endpoint and sends requests directly from the visitor's browser. It does not accept or retain API keys. The included Gradio implementation exposes a `generate` endpoint and launches with MCP support once the Space is switched to Gradio with ZeroGPU hardware.