🖤 Gemma 4 E4B — Agentic Tool Calling Demo

First live demo of Gemma 4 E4B with baked-in Opus reasoning + native tool calling

deadbydawn101/gemma-4-E4B-Agentic-Opus-Reasoning-GeminiCLI-mlx-4bit · Built by RavenX AI

What makes this different: Most HF demos are just chat. This runs a real agentic loop — the model decides when to call tools, executes them, injects results, and reasons to a final answer. Opus reasoning is baked into the weights (<think> tags work without any adapter).


Try these

⚙️ Agent Settings


🔧 Available Tools

  • calculator — math expressions
  • get_datetime — current time / timezone
  • web_search — live web search
  • run_python — execute code snippets

📦 Model

gemma-4-E4B-Agentic-Opus-Reasoning-GeminiCLI-mlx-4bit

10.5 GB · bfloat16 fused
Opus reasoning baked in
No adapter needed


🖤 RavenX AI

GitHub · HuggingFace


🖤 Built by RavenX AIturboquant-mlx · gemini-cli · mlx-gemma4

Opus 4.6 reasoning + Claude Code tool-use fused into Gemma 4 E4B weights


🖤 RavenX Ecosystem

Model  ·  Tool Calling Demo  ·  OpenClaw Sandbox  ·  GitHub  ·  TurboQuant

Built by RavenX AI