Minna is a local-first search database and AI assistant for your documents. Files are indexed on your Mac, they are never uploaded. All AI content is backed by real sources directly from your database.

Named after Minerva the Roman goddess of Wisdom.

Features

• Search with natural language (vector) & exact match (FTS5) through all files added to Minna.
  • Search is blazingly fast ~12 ms for 250 long documents (30,000 total document pieces).
  • Add PDF, Markdown, TXT, HTML, XML, and OPML to the search database.
  • The search index is one portable .irisdb file you can move between machines.
• Ask Minna about anything in any document.
  • Every generated statement is backed by a direct quote in your database.
  • Citations are presented in a sidebar during genreation.
• Multiple LLM providers:
  • Support for MLX on-device models, the standard is Qwen3 4b Instruct.
  • Support for on-device and off-device Ollama.
  • Support for Apple Intelligence, when enabled on the system.
  • Support for Claude through an Anthropic Key.
  • Support for OpenAI through an OpenAI Key.
  • Support for Gemini through a Gemini Key.
• Document Preview:
  • Preview files in-place without leaving Minna.
  • Ask Minna about the previewing document.
    • Same feature set as Ask Minna across all documents, except it is constrained to just one file.
• Organize added files into folders & subfolders.
• File Actions:
  • Change file Title or Description manually.
  • Use Apple Intelligence to generate file descriptions.
  • Assign file colors.
• Sort folders by name & creation date.

Search Experience

https://github.com/user-attachments/assets/d038c2bb-67a2-4760-bc72-8a670a0915a8

Privacy

Minna takes privacy very seriously. We believe that your files are your own, and shouldn't be seen by anyone else. Any conversations with an AI go directly to the provide you chose, with your key. On-device models (MLX & Apple Foundation Models) make no network calls at all. All LLM provider configurations are encrpyted in the macOS keychain. For more information see SECURITY.md.

A note on Telemetry

Minna uses TelemetryDeck for app telemetry. The goal of this telemetry is to get a broad understanding of Minna's userbase, and is never used for tracking. Personally Identifiable Information is not sent to TelemetryDeck, and the remote configurations are set to drop IP addresses. We do not want your data, so we do not store it.

Installation

You can install Minna for macOS 26.0+ by downloading the latest minna.dmg from the releases page.

Sample Data

Minna works best when you provide it with data. To make finding data easier, a sample of Open Course Work can be downloaded here.

Build from Source

Building Minna from source requires macOS 26+, Xcode 26+, and Git LFS. No configuration is needed to build the project. Config.xcconfig ships blank, which turns telemtry off, and sets signing ot local. For more details see CONTRIBUTING.md.

git clone --recursive https://github.com/impel-intelligence/Minna.git
cd Minna
open Minna.xcodeproj

Testing

Tests for Minna can be run from within Xcode. Tests for submodule projects IrisSearch & LookAtMe must be run from their respective Swift Packages.

Architecture

Minna is a fairly standard macOS app. It uses a default Xcode project, with Swift Packages for code isolation. For more information see ARCHITECTURE.md.

These subpackages are:

• IrisSearch (Submodule)
• LookAtMe (Submodule)
• MinnaChat
• DatabaseSchema

Contributing

Contributions to Minna are greatly appreciated! All contributions should include a Developer Certificate of Origin, in git this is equivalent to adding Signed-off-by at the end of your commits.

Before you submit anything, make sure to read CONTRIBUTING.md, CODE_OF_CONDUCT.md, and SECURITY.md. All contributions to MINNA

AI Contributions

AI Contributions are fine, we have and will continue to use AI to write Minna. However, we request that you properly disclose the usage of AI in any code you submit. This helps the maintainers allocate the proper amount of time for code review. All commits that include work from an AI should include a Co-Authored-By trailer on commits.

Please do not remove the AI attributions that are added in code. We believe it is important to track the usage of AI, and it significantly helps maintainers during the review process.

Contact

Minna's development team can be contacted directly at support@tryminna.com. If you have any security concerns, for the safety of our users, please contact us directly instead of posting publicly.

License

Minna is licensed under Apache-2.0. The code license does not cover the Minna name or marks, see TRADEMARK.md for more information.