Gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF Codes Offline

The new release Gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-GGUF provides a local coding and tool-using agent that runs entirely offline. It allows users to run a private artificial intelligence model that can read, reason, and use tools to complete multi-step technical tasks. The model operates efficiently on machines with limited memory, needing only about 4.5 GB of video RAM to function.
Developer Yuxinlu1 created this project to improve upon the previous version by adding a strong focus on agentic tasks. They rebuilt missing reasoning data from scratch to ensure the model could handle complex coding and debugging workloads. The developer focused on making sure the artificial intelligence stays in the loop to solve problems instead of giving up early.
Project capabilities and intended users
- Runs completely offline without cloud APIs.
- Handles multi-step technical coding tasks.
- Operates on small graphics memory footprints.
- Uses native tool protocols for actions.
- Includes verified chain of thought reasoning.
This tool is designed for people who need a local coding assistant that can execute commands and debug code. It benefits individuals who want to keep their data private by running everything on their own hardware. Anyone with a standard graphics card and modest memory can use it for complex technical work without paying for cloud services.
Developer notes and limitations
Developer Yuxinlu1 notes that this version trades a small amount of general knowledge capability for its strong coding and agentic focus. A smaller file size was withheld from this release because it failed stress testing. The developer is already working on a version three to push the capabilities even further while maintaining the small footprint.
"v2 is specialized for coding / terminal / technical-agentic work, and on those (telecom) it dramatically outperforms the base." Source: Hugging Face