Models
Ollama
The default local engine. Install it, pull a model, point 8gent at it if it lives elsewhere.
Set up
Install Ollama from ollama.com, or on macOS:
brew install ollama
ollama servePull a coding model. Pick one that fits your memory:
ollama pull qwen3:14b # about 9 GB
ollama pull qwen3.5:9b # smallerStart 8gent. On first run it lists what Ollama serves and picks from it. Switch with /model.
Ollama on another machine
OLLAMA_HOST=http://gpu-box:11434 8gentA bare host works too (OLLAMA_HOST=gpu-box); 8gent adds port 11434. OLLAMA_BASE_URL is also read, and wins over OLLAMA_HOST.
The System One judge does not follow OLLAMA_HOST: it uses the Ollama on this machine, or EIGHT_DECIDE_OLLAMA_HOST. See System One.
Helper models on Ollama
| Model | For | Pull |
|---|---|---|
| Selene 1 Mini (about 5 GB) | The System One judge | ollama pull hf.co/AtlaAI/Selene-1-Mini-Llama-3.1-8B-Q4_K_M-GGUF |
qwen2.5vl:7b (about 6 GB) | Vision | ollama pull qwen2.5vl:7b, then /vision model qwen2.5vl:7b |
nomic-embed-text | Memory search by meaning | ollama pull nomic-embed-text |
All optional. See Helper and judge models.
If it does not work
| Symptom | Fix |
|---|---|
| "No models found" on first run | ollama list shows nothing: pull a model. Connection refused: run ollama serve |
| A model is missing | ollama pull <name> |
| You do not want Ollama at all | Use llama-server |
8gent doctor checks the engine, the models and the rest of the setup in one go.