Run local models
MiniTavern is a BYOK (bring your own key) client: requests go to the address you configure; it is not locked to one cloud vendor.
“Run local models” means running inference on your own computer, then connecting MiniTavern on your phone over the local network.
This section has two tracks:
| Type | Use | Entry |
|---|---|---|
| Language model deploy | Chat, continue writing, rewrite image prompts | 30-minute quickstart · Tools overview |
| Image model deploy | /imagine, attachment image gen, backgrounds | Image gen overview |
No GPU and no desire to install servers? Start with Free cloud APIs for trial / free quotas.
MiniTavern does not run large language or diffusion models on the phone. Services run on the PC; the App handles character cards, lorebook, chat, and image-gen requests.
Language model deploy
Run an LLM on the PC (Ollama, LM Studio, KoboldCPP, etc.), then in the App choose Custom (OpenAI).
| Item | Suggestion |
|---|---|
| PC | Windows / macOS / Linux; 16GB+ RAM recommended |
| GPU (optional) | NVIDIA / Apple Silicon is faster; CPU-only still works |
| Network | Phone and PC on the same Wi‑Fi (avoid guest/client isolation) |
| Tool | Best for | Default port |
|---|---|---|
| Ollama | CLI simplicity, one-command pull | 11434 |
| LM Studio | GUI, click-to-download | 1234 |
| KoboldCPP | Popular in roleplay communities | 5001 |
| llama.cpp | Run GGUF directly | 8080 |
| LocalAI | Docker / all-in-one compatibility layer | 8080 |
| Jan · text-generation-webui | Other GUIs or plugin ecosystems | See each page |
Suggested order: 30-minute quickstart → Tools overview → Recommended models → Configure in MiniTavern → stuck? Troubleshoot.
Image model deploy
Run Stable Diffusion–family services on the PC. In the App: Settings → Image gen → Image gen link → A1111 or ComfyUI.
| Backend | Default port | Guide |
|---|---|---|
| A1111 / SD WebUI | 7860 | Needs --api --listen |
| ComfyUI | 8188 (Desktop often 8000) | Needs --listen |
Overview and hardware: Image model deploy; filling the link: Connect App.
Important reminders
- The address on the phone must be the PC’s LAN IP — not
localhost/127.0.0.1. - Language-model Base URLs usually end with
/v1(e.g.http://192.168.1.8:11434/v1). - A1111 image gen needs
--api; ComfyUI must be reachable on the LAN. - Allow the ports in the firewall; do not expose an unauthenticated API to the public internet.