Stack · Five models, one toolchain
The stack
Three models run locally, on Tony's own hardware. Two run on Ollama Cloud Pro. Every one of them is called through Ollama, so moving work between local and cloud is a change of model name, not a change of tools.
01 — Models
What each model does
| Model | Role | What it does |
|---|---|---|
| Local Local hardware · Ollama | ||
| qwen3.8:27b | Daily driver | The default model for everyday chat, drafting and questions. If a job doesn't need the cloud, it goes here. |
| qwen3.5:9b | Home Assistant voice | The model behind voice control in Home Assistant. Smaller, so replies come back quickly. |
| embeddinggemma-2 | Notes search | An embedding model. It turns notes into vectors so they can be searched by meaning, not just by keyword. |
| Cloud Ollama Cloud · Pro plan | ||
| glm-5.3:cloud | Hard coding | The heavy hitter, saved for difficult coding problems that need more than the local daily driver. |
| glm-5.3-flash:cloud | Bulk work | The cheap, fast option for high-volume jobs where cost per run matters more than depth. |
02 — How it runs
Same command, either side
# local — on local hardware
$ ollama run qwen3.8:27b
# cloud — Ollama Pro
$ ollama run glm-5.3:cloudWhere things live
Local model files are stored at D:\Ollama\models on Tony's PC. Cloud models run on Ollama's servers under the Pro plan, so they take no local disk space or GPU memory.
API keys and credentials are never committed to this site's code. Anything secret lives in environment variables, entered by Tony.