Gemma 4 Mycelium

The first open-source model in the Crowe Logic family. A Gemma 4 E4B derivative for commercial and at-home mushroom cultivation, distributed as a self-contained Ollama image that runs fully offline on a single laptop.

What it does

Answers cultivation questions the way a senior commercial grower would — with diagnostic discipline. Where general-purpose Gemma 4 will confidently identify a contamination on first glance, Gemma 4 Mycelium asks the questions a working mycologist would ask first: species, inoculation source, visual context. The Modelfile system overlay enforces “never confabulate contamination diagnoses.”

Use

ollama run Mcrowe1210/gemma-4-mycelium-e4b "Why is my agar plate growing fuzzy green colonies near the edge?"

Or via the dedicated CLI:

pip install crowe-mycelium
crowe-mycelium chat

Built on

• Base model: google/gemma-4-e4b (multimodal — language + vision + audio)
• Customization: Crowe Logic identity overlay via Modelfile SYSTEM directive
• Training corpus (Phase 2, LoRA in progress): commercial mycology library — Lion’s Mane Commercial SOP, The Mushroom Grower Vol 1 + Vol 2, Mycelium EI Engine technical docs

Why offline

Commercial mushroom growers operate in places datacenter AI can’t reach: humid fruiting rooms that kill routers, rural greenhouses without coverage, cold rooms in tunnel houses. When a grower opens an agar plate at 4 AM and sees something wrong, the answer they need has to live one foot from their hand — not in a datacenter a thousand miles away.

Gemma 4 Mycelium is the model for that grower.

Source

• CLI + scripts: github.com/MichaelCrowe11/crowe-mycelium-cli (Apache 2.0)
• Model weights: Gemma Terms of Use (ai.google.dev/gemma/terms)

Acknowledgements

Built on Gemma by Google DeepMind. Submitted to the Gemma 4 Good Hackathon.

Cultivation knowledge derived from operational practice at Southwest Mushrooms, a commercial cultivation business in Phoenix, Arizona operated by Michael Crowe.