447 Downloads Updated yesterday
😈 Uncensored Gemma 4 12B (QAT Q4) for Ollama. 👁️ Audio,🎙️, 🛠️Tool Calling, MTP, Rust, Linux, Windows AI Coding Agents & Multimodal Local AI. 🚀
ollama run jikepjikep_16HEX/gemma-4-12b-nightshift-heretic-uncensored-qat-q4
Applications
ollama launch claude --model jikepjikep_16HEX/gemma-4-12b-nightshift-heretic-uncensored-qat-q4
ollama launch codex-app --model jikepjikep_16HEX/gemma-4-12b-nightshift-heretic-uncensored-qat-q4
ollama launch openclaw --model jikepjikep_16HEX/gemma-4-12b-nightshift-heretic-uncensored-qat-q4
ollama launch hermes --model jikepjikep_16HEX/gemma-4-12b-nightshift-heretic-uncensored-qat-q4
ollama launch codex --model jikepjikep_16HEX/gemma-4-12b-nightshift-heretic-uncensored-qat-q4
ollama launch opencode --model jikepjikep_16HEX/gemma-4-12b-nightshift-heretic-uncensored-qat-q4
Models
View all →Readme
👁️🎙️ Gemma-4-12B Nightshift Heretic Uncensored (QAT Q4)
An uncensored Gemma 4 12B Unified Transformer optimized for Ollama, engineered for multimodal AI, native Vision, native Audio, AI Coding Agents, Rust systems programming, Linux development, Tool Calling, Function Calling, and 256K long-context reasoning.
Built upon Google’s revolutionary encoder-free unified multimodal architecture, this Nightshift Heretic edition delivers fast local inference, deterministic software engineering, advanced visual reasoning, and native audio understanding while maintaining an efficient memory footprint suitable for modern laptops and desktop workstations.
Designed for developers who demand high-performance local AI, practical engineering workflows, and maximum productivity without unnecessary conversational overhead.
🚀 Overview
Nightshift Heretic transforms Gemma 4 12B into a powerful local-first multimodal development platform for engineers, researchers, reverse engineers, and AI automation specialists.
Unlike traditional multimodal models that depend on separate vision and audio encoders, Gemma 4 12B integrates text, images, and audio directly into the Transformer backbone. This unified architecture reduces latency, lowers memory consumption, and enables highly efficient multimodal reasoning on consumer hardware.
Optimized for Ollama, GGUF, Rust, Linux, AI Coding Agents, MCP workflows, offline AI, tool calling, and modern software engineering.
⚡ Key Features
- 😈 Uncensored reasoning with minimized refusal behavior
- 👁️ Native Vision understanding
- 🎙️ Native Audio processing
- ⚡ Unified encoder-free Transformer architecture
- 🚀 Multi-Token Prediction (MTP)
- 🛠️ Native Tool Calling
- 🤖 Function Calling support
- 🦀 Optimized for Rust systems programming
- 🐧 Linux, Ubuntu & CLI development
- 🔧 GGUF optimized for Ollama
- 📚 Ultra-long 256K context window
- ⚙️ Deterministic code generation
- 💻 Laptop-ready (16 GB class hardware)
- 🖥 Optimized for local AI inference
🌟 Unified Multimodal Intelligence
Gemma 4 12B processes multiple modalities natively without dedicated external encoders.
Supported workflows include:
- 👁️ Image understanding
- 🎙️ Audio understanding
- 📄 OCR document extraction
- 📷 Screenshot analysis
- 📊 Charts & graphs
- 📐 Technical diagrams
- 🖥 UI inspection
- 📚 Long technical documentation
- 🔍 Reverse engineering
- 🤖 Vision-guided AI agents
💻 Optimized For
- Ollama
- Claude Code
- OpenClaw
- Codex App
- OpenCode
- Hermes
- MCP-compatible workflows
- AI Coding Agents
- Multimodal AI
- Local AI
- Offline AI
- Software Engineering
- Systems Programming
- Terminal Automation
🦀 Programming Languages & Technologies
Primary languages
- Rust
- C
- C++
- Zig
- Python
- Go
- Bash
- PowerShell
- JavaScript
- TypeScript
Developer ecosystem
- Cargo
- Tokio
- Axum
- LLVM
- Docker
- Git
- SQL
- CMake
- Linux APIs
Optimized for
- async programming
- unsafe Rust
- memory safety
- multithreading
- networking
- backend development
- embedded systems
- zero-copy pipelines
- reverse engineering
- performance optimization
⚙️ Native Tool & Multimodal Calling
Supports modern AI agent workflows using native Tool Calling, Function Calling, Vision reasoning, and Audio processing.
Ideal backend for:
- Claude Code
- OpenClaw
- MCP ecosystems
- Autonomous Coding Agents
- Local AI Assistants
- Vision-driven automation
- Audio-enabled workflows
📊 Core Architecture
| Feature | Specification |
|---|---|
| Architecture | Gemma 4 12B Unified Transformer |
| Multimodal | Native Text + Vision + Audio |
| Vision | Encoder-Free |
| Audio | Native |
| Context Window | 256K Tokens |
| Prediction | Multi-Token Prediction (MTP) |
| Runtime | Ollama |
| Format | GGUF |
| Tool Calling | Native |
| Function Calling | Supported |
⚙️ Recommended Configuration
| Parameter | Value |
|---|---|
| Temperature | 0.2 |
| Context | 256K |
| Runtime | Ollama |
| Quantization | QAT Q4 |
| Reasoning | Deterministic |
| Deployment | Local / Offline |
🎯 Best Use Cases
- 👁️ Multimodal AI Coding Assistant
- 🎙️ Audio Transcription & Analysis
- 📷 Screenshot Interpretation
- 📄 OCR & Document Processing
- 🤖 AI Coding Agents
- 🦀 Rust Development
- 🐧 Linux Engineering
- ⚙️ Backend Development
- 🌐 Network Services
- 📡 Infrastructure Automation
- 🔍 Reverse Engineering
- 💾 Embedded Systems
- 🚀 Local AI
- 🔒 Offline AI
- 📚 Long-Context Software Engineering
🔑 Keywords
Ollama, Gemma 4, Gemma 4 12B, Unified Transformer, Encoder-Free AI, Native Vision, Native Audio, Multimodal AI, GGUF, QAT Q4, AI Coding Assistant, AI Coding Agents, Rust, Linux, Ubuntu, Tool Calling, Function Calling, MCP, OCR, Screenshot Analysis, Local AI, Offline AI, Software Engineering, Systems Programming, Reverse Engineering, CLI Development.
🔬 Design Philosophy
Nightshift Heretic is built around one principle: deliver maximum engineering capability on local hardware.
By combining Google’s unified multimodal architecture with deterministic reasoning, native vision, native audio, optimized tool calling, and an uncensored development workflow, this edition is designed for developers who build software, automate infrastructure, analyze technical assets, and create next-generation AI agents.
Built for engineers. Optimized for local AI. Designed for the Night Shift.