Privacy & Local AI

How Privacy Works

River was designed around a clear principle: the only things that leave your machine are the ones you explicitly share.

This page covers how that principle works in practice — what's local, what's cloud, what's encrypted, and how you control all of it.

Two ways to run the AI#

How privacy works depends on how you're running River.

River Cloud (the default)#

River hosts the backend, sync server, and storage. Your entities live in River's databases, encrypted at rest and in transit. AI inference goes to whichever model you've selected (cloud or local).

This is the easiest mode and the one most users start with. Privacy is enforced by encryption, ACLs, and our internal policies.

Local-only mode#

Even on River Cloud, you can configure the AI to use exclusively local Ollama models. In this mode, your prompts never leave your device. Storage still goes to the cloud (encrypted), but inference is local.

What's encrypted#

Data At rest In transit
Entities (docs, mail, tasks…) Yes TLS
Y.js documents (CRDTs) Yes TLS
OAuth tokens (integrations) Encrypted with org-specific key TLS
Connector credentials (per-user) Fernet-encrypted, per-user keys TLS
File uploads Yes TLS
Search index Yes TLS

Connector credentials#

For integrations that store API keys (Stripe Connect, Crisp, custom HTTP tools), each user's credentials are Fernet-encrypted with a per-user key. Even with database access, no one can decrypt one user's credentials with another user's key.

What never leaves your machine#

When you use the desktop app:

  • Files on your computer, unless you upload them or the AI reads them for something you asked
  • River's memory of the files and folders it has opened on your computer
  • Local Ollama inference (entirely on-device)

What does leave your machine#

When you use River Cloud:

  • Entities stored in your Dataspace (encrypted)
  • Edits routed through the WebSocket server for multiplayer (encrypted)
  • AI prompts sent to whichever model you've selected (cloud models route through the provider; local models stay on your hardware)
  • The parts of your computer's files the AI reads while working on your request, which go to that same model. In the desktop app the AI can read and edit any file your computer allows.
  • Pictures of River's screen the AI looks at while it works in River's windows, which also go to that model. The AI sees only River, not the rest of your computer's screen.
  • Integration sync (e.g., Google Drive → River) flows through OAuth + your selected sync path

Data ownership and export#

You own your data. Always.

  • Export any entity, Group, Dataspace, or your entire workspace at any time
  • Delete anything irreversibly
  • Take it elsewhere — entities export as Markdown, JSON, or original file formats

What we never do#

  • Train models on your data
  • Sell or share your data with third parties
  • Read your entities for any purpose other than running the system
  • Use your prompts in our research

This is the explicit, contractual promise.

Audit logs#

For Org Admins and Owners, audit logs (Premium plan) record every privileged action — member additions, permission changes, integration connections, data exports. See Admin Dashboard.

Tips#

  • For a balanced posture: River Cloud + local-only AI mode for sensitive Spaces, cloud AI for the rest.
  • Audit your share list quarterly — Settings → Sharing → Active shares shows everything currently exposed.