Repository navigation
AI Model Setup
Sprigo doesn't ship a language model. Lessons, stories and the tutor are written by a model you choose. Until one is set up, lessons are locked.
You can change the provider any time in Settings → Advanced → AI provider. Test connection checks it and lists the available models.
- Install Ollama and start it.
- Open Sprigo. It finds Ollama at
http://localhost:11434. - If no model is installed yet, Sprigo offers to download
qwen3:8b(about 5.2 GB).
Prefer a model of 8B parameters or more. Smaller models run, but they write noticeably weaker lessons and sometimes wrong answers, so Sprigo warns when you pick one.
To pull a model yourself:
ollama pull qwen3:8b- Install LM Studio and download a model.
- Start its local server (default
http://localhost:1234) and load the model. - In Sprigo, choose LM Studio as the provider.
If your computer can't run a local model well, use a cloud API with your own key:
| Provider | Notes |
|---|---|
| OpenAI | Needs an API key. |
| Anthropic | Needs an API key. |
| Gemini | Needs an API key. |
| OpenRouter | One key for many models. |
Keys are stored in the system keychain (macOS Keychain, Windows Credential Manager, Secret Service on Linux), never in Sprigo's database. With a cloud provider, the lesson prompts and your chat messages go to that provider.
A rough guide for local models:
| Model size | Memory | Experience |
|---|---|---|
| under 8B | 8 GB | Works, weaker lessons |
| 8B | 16 GB | Recommended starting point |
| 14B and up | 32 GB or a GPU with enough VRAM | Better lessons, slower replies |
On Apple silicon, Ollama uses the GPU automatically. Lessons are generated once and cached, so a slower model mostly affects the first visit to a step and the live tutor's reply time.
On your phone (Settings → Use on your phone), Sprigo uses the desktop's Ollama. Cloud providers aren't available there, because the keys stay in the desktop keychain.
Having trouble? See Troubleshooting.
Using Sprigo
Contributing