Privacy Policy
MiniLLM is local-first: models and chats stay on your machine. This policy explains the limited cases where the network or this website is involved.
Local-first design
MiniLLM mainly runs models and the OpenAI-compatible server on your own computer. Prompts, chat history, images you send, and model outputs are processed by llama-server locally and are not uploaded to Bashige.
Data we do not collect
- No account, no registration, no telemetry/analytics SDK in the app.
- No upload of your conversations, prompts, or images to any Bashige server.
- No sale of personal data. The desktop app ships no ad SDKs and this website shows no ads — see Cookies & advertising.
- The official website does not require login and does not intentionally collect personal information beyond ordinary web-server request logs hosted by your provider.
When the network is used
Network access occurs only when you ask the app to:
| Trigger | Purpose | Data sent |
|---|---|---|
| Install / Update | Download llama.cpp builds (CPU/CUDA/Vulkan/SYCL/ROCm/OpenVINO/ARM64) | Normal HTTP(S) GET request |
| Certificate refresh | Refresh root certificates on new PCs | Certificate download only |
| Model download | If you fetch GGUF/mmproj from external sources | Per the model host's logs |
llama.cpp or models requires Internet access; everything else stays local.Local configuration
Settings are stored on your device at:
%APPDATA%/MiniLLM/config.json
It remembers installed version, model directory, model file, server parameters (host, port, n_ctx, n_gpu_layers, threads, slots, API key if you set one), and sampling parameters. You can delete this file to clear all app state.
Website & downloads
- Official domain: https://minillm.bashige.com/
- Downloads are static files served over HTTPS; your hosting provider may keep standard access logs (IP, time, path, user-agent) per their own policy.
- If you email bashigestudio@hotmail.com, that address and message content are used only to reply to you.
- Prefer official builds only; see Terms §4 Official downloads.
LAN & API security
- Default host
127.0.0.1→ reachable only from this PC (safest). - If you set
0.0.0.0, anyone who can reach your port may call the API — always set API key and allow only the chosen TCP port in Windows Firewall. - Never port-forward the llama-server to the public Internet without authentication and TLS termination.
Third-party licenses
MiniLLM may download or invoke third-party open-source components such as llama.cpp and does not bundle model weights. Those projects and any models you choose are governed by their own privacy and license terms — see Third-party licenses.
Your choices
- Stay offline after install: llama.cpp and models already on disk work without further downloads.
- Delete
%APPDATA%/MiniLLM/config.jsonto erase saved settings. - Uninstall llama.cpp from the Install tab; remove MiniLLM.exe (it is portable, no registry residue by design).
- Use
127.0.0.1only, to keep the API private.
Contact & changes
Questions about this policy: bashigestudio@hotmail.com. We may update this page when the product changes; the effective date above will be revised. Continued use after an update constitutes acceptance of the revised policy.
Cookies & advertising
This policy covers both the MiniLLM app and this website. Neither shows ads:
- This website shows no ads and uses no advertising cookies.
- The MiniLLM desktop app itself does not load ads and does not use advertising cookies.
- We do not sell personal data.