Privacy Security

OpenLoomi gives you complete control over your data:

• Access Controls: Choose exactly which platforms and data OpenLoomi can access
• Data Retention: Configure how long OpenLoomi retains processed data
• Delete Anytime: Delete all your data at any time with one click
• Export: Export all your data in standard formats

We believe privacy and intelligence aren't mutually exclusive. You should never have to choose between powerful AI capabilities and data control. OpenLoomi delivers both.

🔐 Our Privacy Principles

Local-First Architecture

Your original messages and files stay on your device. OpenLoomi only accesses the minimum data needed to complete tasks—no unnecessary uploads.

• On the desktop app, raw data never leaves your local environment
• Only processed results are transmitted when needed
• Complete control over what data OpenLoomi can access

Why Local-First?

OpenLoomi is built on the principle that privacy should not be a trade-off. Traditional AI assistants send your data to remote servers for processing—every conversation, every file, every query becomes训练 data or at minimum passes through someone else's infrastructure.

OpenLoomi flips this model. We believe your personal AI agent should work for you, not the other way around. By processing as much as possible locally, we ensure:

• Your data never leaves your control — Messages, files, and context are processed on your machine
• No dependency on cloud services — Work offline or in low-connectivity environments
• Faster response times — Local inference eliminates network round-trips for many tasks
• Cost efficiency — Run your own models without paying per-token fees

What Stays Local?

OpenLoomi maintains a local knowledge graph on your device that powers your personal memory and context:

Data directory path by platform:

• macOS: ~/.openloomi/data
• Linux: ~/.openloomi/data
• Windows: %USERPROFILE%\.openloomi\data

All data at rest is encrypted. The database is stored in SQLite with application-level encryption, ensuring that even if your disk is accessed directly, your personal data remains protected.

Local Scenarios

OpenLoomi excels in these fully local scenarios:

Offline Email Drafting When you're traveling without internet, OpenLoomi can draft replies using your local email context and cached data. Queued replies are sent automatically when connectivity returns.

Air-Gapped Document Q&A Ask questions about sensitive documents stored on your local machine—no document content ever leaves your device. Only the question and extracted answer travel across the network.

Private Voice Notes Record voice memos that are transcribed and summarized entirely locally. Your voice data never touches a cloud STT service.

Local RAG (Retrieval-Augmented Generation) Build a personal knowledge base from local files (PDFs, Markdown, Notion exports). OpenLoomi indexes and retrieves from your local filesystem—nothing is uploaded to external vector databases.

Configuring Local Models

All local model settings are managed in Settings → API Settings, which covers conversation models and embedding models.

Conversation Models (Language Models)

The Conversation models section lets you configure an Anthropic-compatible provider. This is the model used for chat, reasoning, and general task execution.

To configure a language model:

1. Go to Settings → API Settings → Conversation models
2. Find the Anthropic compatible provider card
3. Fill in the fields:
   • API Key: Your API key (e.g., sk-ant-...)
   • Base URL: The endpoint URL (e.g., https://api.anthropic.com for Anthropic, or a custom endpoint like http://localhost:11434 for a local Ollama server)
   • Model: The model ID to use (e.g., claude-sonnet-4-6, or your local model name)
4. Toggle Enabled to activate the provider
5. Click Test to verify the connection works
6. Click Save to persist your settings

You can also point it to any Anthropic-compatible endpoint (e.g., a local Ollama server, a proxy service, or a custom deployment) by changing the Base URL. If your workspace has system defaults, they are shown as reference—your personal settings override them.

Embedding Models

The Embedding models section controls how OpenLoomi creates vector embeddings for knowledge base, memory, and semantic search. You can choose between Online API (cloud) or Local model.

To switch to local embeddings:

1. Go to Settings → API Settings → Embedding models
2. Choose Local model instead of Online API
3. Select a model from the dropdown:
   • all-MiniLM-L6-v2 (384 dimensions) — lightweight, fast
   • bge-large-zh-v1.5 (1024 dimensions) — higher quality, bilingual support
   • Custom model ID or local path — enter any Hugging Face model ID or local path
4. Optionally enable Use local files only to restrict to model files already on your device (disables automatic downloads)
5. Click Test to verify — the first test may download the model and take longer
6. Click Save to persist

To go back to cloud embeddings, choose Online API and fill in your API Key, Base URL, and Model ID (e.g., text-embedding-3-small).

Speech Models (STT / TTS)

Voice transcription uses the Online API (cloud) endpoint configured in Embedding models settings. For fully local voice transcription, a local STT configuration panel is planned — stay tuned.

💡 Privacy Tip: Set the Base URL to your local Ollama server (http://localhost:11434) to route all inference through your own machine. Your prompts and context never leave your device.

End-to-End Encryption

All authorized data is encrypted with AES-256 industry standard, processed in hardware-isolated trusted execution environments—completely inaccessible to external parties.

• AES-256 encryption for all stored data
• Hardware-isolated trusted execution environments
• Zero visibility into your raw data

Zero Training Commitment

Your data will never be used to train public AI models—now or ever. In writing, legally guaranteed.

• Explicit legal commitment in our terms of service
• No use of your data for model training
• Full transparency on data usage

Auditable Compliance

All critical operations are logged in tamper-proof audit logs. SOC 2 compliance certification in progress.

• Complete audit trail of all data access
• Enterprise-grade security standards

You can view exactly when and why OpenLoomi accessed your data. Every access is logged with:

• Time: When the data was accessed
• Platform: Which platform (Telegram, WhatsApp, Gmail, etc.)
• Data Type: What kind of data was accessed
• Reason: Why the access was needed (e.g., "Generating insights", "Replying to messages")

💡 Transparency First: You have full visibility into all data access. If you notice any suspicious activity, you can revoke access immediately from Settings.