Skip to main content

Privacy & Security

OpenHuman is designed so that the memory of your life lives on your machine. The local SQLite Memory Tree, the Markdown Obsidian vault, your audio buffers, all of that stays under your control.

Privacy by Design​

The Memory Tree is local. The SQLite database and the Markdown vault live on your machine. The agent reads from them locally; nothing about your raw source data sits on the OpenHuman backend.

Integration tokens are held by the backend. OAuth tokens are never written to disk in plaintext on your device. The OpenHuman backend brokers each integration request.

No training on your data. Your conversations, your Memory Tree, and your personal information are never used to train AI models.

What stays on your machine​

Memory Tree SQLite databaseLocal - <workspace>/memory_tree/chunks.db
Obsidian Markdown vaultLocal - <workspace>/wiki/
Audio capture buffersLocal. Discarded after STT.
Local model stateLocal.

What the backend handles​

LLM callsProxied through the backend under one subscription.
Web search proxyUses backend proxy by default.
Integration OAuthToken storage and rate limiting for 118+ integrations.
TTS streamingGenerated and discarded - not retained.

Permissions and access control​

  • OpenHuman accesses an integration only after you complete its OAuth flow.
  • Each connection has its own scope; you can revoke at any time from the Skills tab.
  • Auto-fetch runs continuously while connected, but bound by OAuth scope, sync interval, and daily budget.

Security​

  • Encrypted in transit. All communication uses TLS.
  • Key in keyring. Local secrets use OS-level credential storage.
  • Sandboxed skills. Each skill runs in its own isolated environment.
  • Workspace-scoped tools. Filesystem tools operate within the workspace boundary.
  • Short-lived tokens. Authentication tokens are time-limited.

See also​