AI File Sorter: LLM-Driven File Renaming With a Review Table Before Anything Moves
Cross-platform desktop application for content-aware file organization and renaming. Supports local and remote LLMs, preview-based workflows, and fully user-controlled changes.
At a glance
- What is it?
- AI File Sorter is a C++ desktop application that proposes categories and filenames for images, documents and media files using local or remote LLMs. Its defining constraint is that nothing is moved or renamed until you approve the suggestions, and the README points to a single-step undo when you do.
- Who is it for?
- Adopt AI File Sorter if you want a desktop tool that proposes names and categories for a folder of images, documents or media files and shows you the proposal before committing, and if you are willing to run a local model or supply your own API key. Do not adopt it if you need unattended, scheduled organization of a live directory tree, because the README's workflow is built around a review table and explicit approval.
- Can I use it commercially?
- Yes, with strict conditions. AGPL-3.0 is a network copyleft licence: if people use a modified version over a network, for example as a hosted service, you must offer them its source code under the same licence.
- Is it still maintained?
- Yes. The repository received new commits within the last day.
- What is it written in?
- Mainly C++, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 15, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
The problem: filenames that carry no information
A camera writes IMG_2048.jpg. A scanner writes scan_0012.pdf. A download folder fills with files whose names describe nothing about their contents, and the only way to find anything later is to open files one at a time. AI File Sorter targets that specific gap. According to the README, it analyzes picture files with built-in visual LLM backends and suggests names such as clouds_over_lake.jpg in place of IMG_2048.jpg, analyzes supported document files and proposes names based on their text content, and cleans up audio and video filenames using metadata already stored inside those files. The README gives the example of a suggestion like 2024_artist_album_title.mp3 built from year, artist, album and title tags. The intended users are people tidying Downloads folders, external drives or NAS storage, where the README says the app groups files based on names, file types, folder context and past sorting results. This is a personal-archive and ad-hoc-cleanup tool, not a records-management system with retention rules.
What actually happens between selecting a folder and moving a file
The README describes four steps: point the app at a folder or drive; files and, where applicable, image content are analyzed using the selected local or remote model; category and rename suggestions are generated; you review and adjust before anything is changed. Categories and optional subcategories are suggested per file, and rename suggestions are added for supported file types. Only after confirmation does the app create the required folders and sort the files. Two mechanisms sit on top of that loop. The first is a categorization cache with learned behavior: the README says the app combines AI suggestions with optional whitelists, recent similar results and approved review decisions, so consistency improves as you accept or reject suggestions. The second is the undo path, Edit -> Undo last run, which the README offers as the reversal route after changes have been applied. The whitelist and cache are the parts worth understanding before you rely on the tool, because they mean output is not purely a function of the current folder contents; earlier decisions feed later ones. The README does not spell out the cache format or where it is stored, so treat that as something to inspect yourself if consistency across runs matters to you.
Model backends, API keys and the local-versus-remote choice
The README's privacy section states that AI File Sorter can run entirely on your device, that files, filenames, images and metadata stay on the computer when a local model is used, and that no telemetry is sent. An internet connection is needed only if you choose a remote model. The repository's platform logos cover Vulkan, CUDA, Apple Metal, Windows, macOS and Linux, which indicates the local inference path is meant to work across GPU APIs rather than a single vendor. For remote use, the README documents three routes: your own OpenAI API key, your own Gemini API key, and a custom OpenAI-compatible API. That third option is the one that matters for anyone running a self-hosted inference server, since an OpenAI-compatible endpoint is a common interface. Document analysis uses a text LLM, image analysis uses a visual LLM, and the README lists required visual LLM files as a separate topic, which suggests local visual analysis needs model files placed in a known location rather than downloading them transparently. The README also mentions a system compatibility check and a benchmark dialog, shown in a macOS screenshot, so the app appears to test whether your hardware can run a given local backend before you commit to it.
Getting it running: installers, then a small test folder
The README lists installation sections for Linux, macOS and Windows, and links downloads from SourceForge and the Microsoft Store. The project's own site is filesorter.app. The recommended first run is explicit and worth following: copy 20 to 50 files from Downloads, screenshots, photos or documents into a temporary folder, run the analysis, and inspect the review table before applying anything. The README frames this as keeping the first run low risk, noting that the AI only suggests categories and filenames and that no move or rename happens until approval. There is also a testing section in the README, including optional headless live LLM tests, which is relevant if you intend to build from source rather than use a packaged build. Beyond the model-selection options in the main window and the API key settings for OpenAI, Gemini and custom OpenAI-compatible endpoints, the README does not document command-line flags or a configuration file format in the material available here. If you need scripted or reproducible runs, that absence is a real constraint, not a detail to work around later.
Where the review-table design becomes the wrong tool
The approval step is the product's safety property and also its ceiling. A workflow that requires a human to read a review table and confirm does not scale to a directory that changes continuously, and the README describes no watcher, daemon or scheduled mode. The same applies to large archives: the README's own advice is to start with 20 to 50 files rather than a full archive or drive, which is a reasonable caution but also an admission that first-run behavior on a large tree is not something the documentation promises. The undo mechanism is scoped to the last run, so a sequence of accepted runs does not have a documented multi-step rollback. Category and rename suggestions are also only as good as the model behind them; the README does not publish accuracy figures, and the repository's download badges are not evidence of suggestion quality. Finally, the app targets images, documents and supported audio and video formats. Files outside those categories get category suggestions based on names, types and folder context, but no content-derived rename. If your clutter is mostly arbitrary binary files, the rename half of the tool does little for you.
How it differs from a rule-based renamer
The obvious alternative is a deterministic batch renamer driven by patterns, EXIF fields or media tags, of which there are many. The difference is where the decision comes from. A rule-based tool applies a pattern you wrote, so a file named IMG_2048.jpg becomes something like 2024-05-01_001.jpg only if the metadata supports it, and a file with no useful metadata stays opaque. AI File Sorter instead asks a model to look at the image content or the document text and propose a semantic name, which is the only way to get clouds_over_lake.jpg from a file whose EXIF data says nothing about clouds. The trade is determinism for meaning. A rule-based renamer gives the same output for the same input every time and can be run unattended; AI File Sorter gives a suggestion that varies with the model and with the app's accumulated review history, and it expects you to look at it. The README's mention of audio and video metadata handling shows the two approaches overlap there, since tag-derived names are exactly what a rule-based tool would produce. Choose based on whether your problem is missing metadata or missing meaning.
Licence, maintenance and what to check before depending on it
AI File Sorter is licensed under AGPL-3.0. For individual desktop use that is unremarkable. If you are considering embedding it in a product, or offering a modified version as a network service, the AGPL's source-availability conditions apply to modified versions and to network use, and that is a question for your own legal review rather than something to settle from a README. On maintenance, the repository shows v1.9.0, v1.9.1 and v1.9.2 released within roughly ten days in August 2026, with the last push to the default branch in September 2026, so the project is actively changing. Frequent point releases during a period like that can mean fast bug fixes or a moving target, and the material here does not distinguish between the two. The practical upgrade cost is the local model files: if a release changes which backends or model files are expected, you may need to re-fetch them, and the README treats required visual LLM files as a setup step rather than an automatic download. Before depending on it for an archive you care about, verify three things on your own machine: that the compatibility benchmark passes for your GPU backend, that a test run on 20 to 50 copied files produces suggestions you would accept, and that Edit -> Undo last run restores the original names and locations.
Editorial conclusion
Adopt AI File Sorter if you want a desktop tool that proposes names and categories for a folder of images, documents or media files and shows you the proposal before committing, and if you are willing to run a local model or supply your own API key. Do not adopt it if you need unattended, scheduled organization of a live directory tree, because the README's workflow is built around a review table and explicit approval. Before trusting it on a real archive, copy 20 to 50 files into a temporary folder, run the analysis, and check the review table; then confirm that Edit -> Undo last run actually reverses the result on your platform.
Community notes