Privacy policy
SharpMind Studio is a desktop application. It runs on your machine, keeps your models and work on your machine, and contacts a service only when you use a feature that needs one. There is no SharpMind account, no telemetry and no analytics.
The short version
- Chats, projects, models, datasets, training output and settings are stored locally, under your Windows profile.
- Prompts and attachments leave the machine only when you send them to a remote model provider that you configured.
- API keys are stored encrypted and are sent only to the provider they belong to.
- Studio does not phone home, report crashes, or measure usage.
- Uninstalling does not delete your data; removing the data folder is your decision.
What stays on your machine
Everything Studio creates lives under %USERPROFILE%\SharpMind (the same tree the SharpMind command-line tools use). In the Docker image it is the data volume you mount.
| Folder | What it holds |
|---|---|
| Models | GGUF and .smm model files you downloaded or built. |
| Corpus | Datasets used by Data Factory. |
| Training | LoRA runs, checkpoints, exports and Data Factory output. |
| Chat Sessions | Chat transcripts, including attachments you added. |
| Studio | Settings, inference profiles, provider configuration, project registry and gateway state. |
| Workspaces | Work sessions tied to a project folder. |
| Compute | Saved compute hosts, SSH known-hosts pins and deployed-instance bookkeeping. |
Work runs tools against folders you open, so project files stay wherever you keep them. Nothing in this tree is uploaded by Studio itself.
What leaves your machine
Each of these happens only as a direct result of something you chose, and each destination has its own privacy policy and retention.
| Destination | When | What is sent |
|---|---|---|
| OpenRouter, EuroRouter, OpenCode Go | You select a remote model and send a message | The prompt, system and agent messages, and any attachments you added, plus your API key for that provider. Guardrails and retention are theirs. |
| HuggingFace | You download a model or tokenizer | A file request, and your HuggingFace token if you saved one for gated repositories. |
| Vultr, Runpod | You deploy, list or destroy an instance | Your API token and the instance settings in your compute profile. |
| Your own compute hosts | You deploy or run a job | SSH traffic, the container image pull, and the setup scripts run on that instance. |
| api.ipify.org | The gateway needs this machine's public address | The request itself. It returns your public IP; nothing else is sent. |
| GitHub | You open a release or download link | A normal web request, handled by your browser. |
Provider referral links in Settings (runpod.io, eurouter.ai, vast.ai, novita.ai) are ordinary links. They open in your browser and carry the provider's referral parameters, not your data.
Keys and secrets
- Provider, HuggingFace and compute credentials are encrypted with AES-GCM.
- On Windows the encryption key is wrapped with DPAPI for the current user, so copying the profile to another account or machine does not expose the secrets.
- On Linux and in Docker the key file is kept with owner-only permissions instead.
- A saved key is sent only to the provider it belongs to. Removing the key in Settings deletes it locally.
- Pairing tokens are stored hashed, and PINs are salted and deliberately slow to guess.
Pairing and your network
- Studio listens on 127.0.0.1:5117; loopback peers are treated as you.
- The LAN listener on 5118 refuses non-loopback browsers until you enable pairing and approve a device.
- Paired devices get a narrower mobile scope, and can be revoked at any time.
- In 1.1.2 pairing is LAN only. There is no relay and no SharpMind-operated server in the path.
Tools, plugins and MCP
Work tools, embedded plugins and MCP servers run on your machine with your permissions. Studio decides whether a tool may be called and shows you the exact command before it runs, but a plugin or MCP server is its own program: if it talks to the network, that traffic is its own and follows whatever it does, not this policy. Only add tools and servers you trust, and keep their permissions at Off unless you need them.
Microsoft Store version
The Store build is the same application, packaged as MSIX. Microsoft may collect installation and reliability data for Store apps under the Microsoft Privacy Statement. SharpMind adds nothing to that: the app itself still sends no telemetry.
Your choices
- Delete everything: close Studio and remove %USERPROFILE%\SharpMind. Models, chats, training runs and keys go with it.
- Remove one provider: delete its key in Settings → Providers. Its configuration entry can go too.
- Stop remote traffic: use a local model. Nothing leaves the machine.
- Revoke access: remove paired devices, or turn LAN pairing off.
- Uninstall: removes the program, not your data.
Contact
Questions about this policy, or a data-handling concern, go to github.com/Integral2u/SharpMindStudio/issues. This page describes SharpMind Studio 1.1.2; if a later version changes what leaves the machine, this page changes with it.