Gemma: A Five-Layer Cognitive Architecture for Local LLMs
An experimental five-layer cognitive architecture that wraps a local LLM (via Ollama) in persistent memory, self-monitoring, background reflection, and a continuously updated self-model. Developed and tested on Google's Gemma 4 26B for best results, but model-agnostic — any Ollama-compatible model works. The base model is unmodified — the architecture adds continuity, error correction, and self-regulation through external processing layers. Runs entirely offline on consumer hardware. Includes full source code (Python), web interface, and example configuration.
Resource information
- Format
- zip
- Original repository
- 19471965
- DOI
- 10.5281/zenodo.19471965
- Credit basis
- Creators in Zenodo record metadata
- Access
- Open files available from the original Zenodo deposit
- Last checked
- Nov 5, 2026
- License
- cc-by-nc-4.0
- Last record update
- Nov 5, 2026
About this resource
An experimental five-layer cognitive architecture that wraps a local LLM (via Ollama) in persistent memory, self-monitoring, background reflection, and a continuously updated self-model. Developed and tested on Google's Gemma 4 26B for best results, but model-agnostic — any Ollama-compatible model works. The base model is unmodified — the architecture adds continuity, error correction, and self-regulation through external processing layers. Runs entirely offline on consumer hardware. Includes full source code (Python), web interface, and example configuration.