StratoSort Review: Local AI File Organizer for Mac, Windows and Linux

StratoSort installs like any desktop app, so no terminal is needed. Download the installer for your system from the GitHub release page; the AI models are downloaded inside the app after the first launch.

StratoSort is a file organizer that reads content, not just file names. It looks inside documents, spreadsheets, images, scans, audio and video, suggests a name and a folder, and moves files into "Smart Folders" that you describe in plain language.

The repository is named StratoSortStack and describes itself as the successor of the original StratoSort; the website and this review use the name StratoSort. Beyond sorting, the app has a search, chat and graph side that the maker calls Knowledge OS and describes as about half of what it does, plus an opt-in local MCP server.

This review draws on the website, documentation, privacy policy and GitHub repository. It involved no hands-on testing, so how well the app names and sorts files, and how fast it runs, is not rated here.

Setup is an install, a model download and a folder choice. The steps follow the release notes and documentation.

The feature set centres on organizing and finding files. Everything below comes from the website, documentation, release notes and the maker's comments of 10 November 2026.

StratoSort ships two model profiles, both as GGUF files. The Base profile uses Llama-3.2-3B-Instruct for text, llava-v1.6-mistral-7b for vision and all-MiniLM-L6-v2 for embeddings, about 7 GB in total. The Better Quality profile swaps in Qwen2.5-7B-Instruct and nomic-embed-text-v1.5, about 10 GB in total.

The documentation states 8 GB of RAM as the minimum and suggests 16 GB of RAM and an 8 GB GPU for Better Quality. A GPU is optional: the app can use NVIDIA CUDA, Apple Metal or Vulkan and falls back to the CPU. The Whisper speech model (about 150 MB) is downloaded only when you install it. No speed or accuracy figures were reproduced here.

The privacy policy says the app runs on your machine and that document content, file paths, embeddings and search queries do not leave it in the standard configuration. It also says the app includes no analytics, telemetry, crash reporters or feature-flag services that phone home, and the website has no account or sync service.

The network documentation lists the connections the app can make: model downloads from huggingface.co, a check for new releases on GitHub, and a one-off download of the speech model. The maker says the vision runtime and OCR language data ship with the installers and are downloaded only as a fallback if a bundled file is missing, so in normal use the speech model is the only one-off download of those extras, and only when you choose to install it. The release check is on by default in installed builds and can be switched off in Settings. Cloud AI stays off unless you add your own Anthropic or OpenAI-compatible key, and then the text you send goes to that provider under its own terms. The code is public and PromptQuorum did not examine network traffic, so these statements are the maker's own.

The licence is PolyForm Noncommercial 1.0.0, which is source-available rather than open source. Personal and other noncommercial use is free with every feature; commercial use needs a paid per-seat licence from Kennedy Applied Sciences, Inc., quoted by email with no published price.

These are the closest tools in the directory: one that also proposes folders and file names, one that searches and chats over local files, and one general document-chat workspace.

StratoSort is a good fit if you want AI to suggest names and folders for a messy pile of documents, scans and images on a Mac, Windows or Linux computer, with models that, per the documentation, run on the machine and ask for your approval first. It is free for personal use and shipped a release this month.

The caveats are the licence, the large first download, a project that went public recently and claims PromptQuorum could not check. Try it on a copy of one folder before pointing it at your archive.