Minna is published as open source specifically so that its privacy and security behavior can be verified. Reports are welcome and taken seriously.

Please do not open a public issue.

Report privately through either channel:

• GitHub private security advisories - preferred, since it keeps discussion and the eventual fix in one place.
• Email support@tryminna.com with SECURITY in the subject line.

Please include:

• What the issue is and why you believe it is a security problem
• Steps to reproduce, ideally with a minimal case
• The Minna version (Minna -> About) and your macOS version
• Whether you intend to disclose publicly, and on what timeline

• Acknowledgement within 3 business days.
• An assessment and a rough remediation timeline within 10 business days.
• Credit in the release notes and the advisory, unless you prefer otherwise.

We ask that you give us a reasonable opportunity to ship a fix before disclosing publicly. We will keep you informed of progress and will not ask for open-ended delays.

Minna is a local-first macOS application. The following are in scope:

• Unintended exposure of indexed document contents, chat history, or the .irisdb index
• Mishandling of user-supplied AI provider API keys, which are stored in the macOS Keychain
• Telemetry or network traffic that transmits more than what is documented below
• Sandbox escapes, privilege escalation, or code execution via crafted documents during indexing
• Weaknesses in the Sparkle auto-update path, including update signature verification

Out of scope:

• Vulnerabilities in third-party AI providers (Anthropic, OpenAI, Google, Ollama) themselves - report those to the provider
• Issues that require an already-compromised machine or physical access to an unlocked device
• Findings from automated scanners without a demonstrated impact on Minna

Stated here so that reports can be measured against intended behavior. If you observe traffic that contradicts this, that is a reportable issue.

• Your documents and chats stay on your machine. They are indexed locally into an .irisdb package and are never uploaded by Minna.
• AI provider requests go directly from your machine to whichever provider you configured, using your own API key. On-device models (MLX, Apple Foundation Models) make no network requests at all.
• Model downloads come from cdn.tryminna.com.
• Update checks (direct-download builds only) come from the Sparkle appcast.
• Crash reporting and analytics are opt-in at build time and are disabled in builds from source. In official release builds they collect aggregate counts and a model name - no document contents, no chat contents, and no personally identifying information (sendDefaultPii is off). See Minna/TelemetryWrapper.swift, which is deliberately small enough to read in full.

Security fixes are issued for the latest released version. Please update to the current release before reporting.