Architecture: Air-Gapped by Design

Native

Intelligence

Minna is engineered to solve AI Amnesia without compromising your digital perimeter. We provide the Signal - you own the Collection.

Data Privacy

The Three Pillars of Privacy

🏠

Local-First Ingestion

Data from connected sources (Slack, GitHub, Linear) is pulled directly to your Base Station (Mac). Minna does not route your private messages or code content through our cloud servers.

📁

The Restricted Collection

~/.minna

Your vector embeddings and personal history reside in a restricted local directory, protected by OS-level permissions (0o600). Your record is invisible to us.

🔗

The Fast-Path Bridge

We provide a "Zero-Trust" handshake. We store only the encrypted OAuth tokens required to maintain source connections. Your local Signal Engine does the heavy lifting.

Data Handling

How We Handle External Sources

🔍

Least Privilege

We request the minimum scopes necessary (e.g., history:read) to build your local Record. Read-only where possible. No admin access ever.

Ephemeral Transit

Data fetched from third-party APIs is ephemeral in transit and exists permanently only on your local hardware. Minna does not maintain a secondary cloud database of your source content.

🚫

No Training Data

We never use your institutional memory, codebases, or conversations to train LLMs. Your data is your competitive advantage.

Security & Compliance

Audit-Ready by Design

Built for the security-conscious. Every architectural decision prioritizes your data privacy.

🔒

Transport

TLS 1.2+

All API communication uses modern encryption protocols

🔐

Storage

Secure Enclave

Sensitive tokens are stored in the macOS Keychain with biometric protection

🗑️

Deletion

User Controlled

Uninstalling the Base Station and clearing ~/.minna removes all indexed data