What is embly?
A local-first AI file organizer — extract, classify, embed, file.
The problem
Most people accumulate files faster than they file them. Downloads pile up, desktops become dumping grounds, and the document you need is somewhere in a folder you named "misc" three years ago. Traditional file managers ask you to do the organizing yourself: drag, rename, sort, repeat. Cloud-based search tools send your files to a server and return keyword matches. Neither approach scales to the reality of how people actually work — across hundreds of thousands of files, with inconsistent naming, across every file type imaginable.
How embly works
When you drop a file into embly, four things happen immediately on your device. First, embly extracts the content: text is pulled from PDFs and documents, images are passed through an OCR and vision model, audio and video are transcribed using Whisper. Second, the content is classified — embly assigns a category and subcategory to the file using a fine-tuned classifier trained on a broad corpus of real-world files. Third, the content is embedded: a vector representation is generated and stored in a local index. Fourth, the file is filed — moved to a suggested location in your file system, or left in place if you prefer. Once a file is indexed, you can find it with a plain English query like "the invoice from the plumber last spring" or "my notes on the protein folding paper". Queries are resolved against the local vector index without any network request.
What embly can read
embly processes a wide range of file types. Documents: PDF, Microsoft Word (DOCX), RTF, EPUB, plain text, Markdown, HTML. Data: CSV, JSON, XML. Code: most programming languages, config files, shell scripts. Images: JPEG, PNG, TIFF, WebP — OCR for text extraction, vision model for scene and content description. Audio: MP3, WAV, M4A, FLAC — transcribed using a local Whisper model. Video: MP4, MOV, MKV — audio track extracted and transcribed. The same pipeline applies to all file types: extract, classify, embed, file.
Local-first by design
Every part of the embly pipeline runs on your device. The classifier, the embedding model, the vector index, and the search engine are all local. No file content, no file name, and no query ever leaves your machine. This is a deliberate architectural decision, not a marketing claim. It means embly works fully offline, on a plane or in a bunker. It means the latency of a search is bounded by your hardware, not a network round-trip. It means you are not subject to the privacy policy of a cloud provider. The CLI and local engine are free and will remain free. A hosted sync layer for teams is planned as an optional paid add-on — but the core product will always run independently of any server.
Who it's for
embly is useful for anyone who accumulates files faster than they can manually organize them. Researchers who download dozens of papers a week and need to find the one about a specific methodology six months later. Developers who collect snippets, configs, and documentation alongside their code. Freelancers who manage contracts, invoices, and client assets across multiple projects. Writers who archive drafts, reference material, and research notes. If you have ever spent more than five minutes looking for a file you know you have, embly is for you. embly is not a replacement for purpose-built apps like reference managers or note-taking tools — it is a layer underneath everything, making your file system navigable by meaning rather than location.
Current status
embly is in early access. The classification engine has been benchmarked on a dataset of 40,000 files drawn from real-world sources, achieving 91% top-1 accuracy and 97% top-3 accuracy across a broad taxonomy of file types and content categories. The local CLI and engine are under active development. A desktop application with a visual interface is in progress. The waitlist is open — early access users will receive the CLI first, followed by the desktop app. Feedback from early users directly shapes the classification taxonomy, the filing logic, and the search interface.