Frequently asked questions

FileMind is a desktop application for Windows 10 or later that organizes folders of research-paper PDFs. It reads each PDF, looks up its DOI or arXiv record, proposes a filename in the form Author_Year_Title.pdf with the evidence behind it, and renames only the files you approve. It then indexes the full text so you can search by keyword or meaning and ask questions that are answered only from your own papers, with page-level citations. Everything runs on your machine.

Researchers, graduate students and academics who have accumulated a large folder of downloaded papers with unhelpful names, and want them renamed, searchable and questionable without uploading anything or building a reference library by hand first.

No. FileMind renames files in place. Your PDFs stay in the folders they are in, under the new names, and nothing is copied into a proprietary store. The search index and metadata live in a single SQLite database file that you can back up, inspect or delete.

No. Filenames are always composed from a fixed template using validated metadata fields. Metadata comes first from DOI and arXiv lookups (Crossref and the arXiv API), then from parsing the document layout, and only as a last resort from a language model extracting structured fields from the text. The language model never writes the filename itself, and every proposal shows its confidence score and the reasons behind it.

Yes. Every applied batch is written to an action journal before any file is touched, and Batch History has a one-click Rollback that restores every file in the batch to its previous name. If FileMind is interrupted mid-batch, it rolls the batch back automatically on the next launch.

Not for the core work. Text extraction, OCR, indexing, search, related papers and the built-in AI model all run locally, and the search and OCR models ship inside the installer. Internet is used to look up DOI and arXiv metadata during a scan, to download the built-in AI model once during setup, and to reach a cloud AI provider if you choose one instead of a local model.

With the built-in model, Ollama or LM Studio: nothing. If you configure a cloud provider (OpenAI, Claude or Gemini), only the short text snippets needed for the current request are sent, never the full PDF. FileMind has no account, no telemetry and no server of its own. DOI and arXiv lookups send the identifier or title to Crossref or arXiv.

Yes. FileMind can push papers directly to Zotero, browsing your collections and validating the connection first, and it exports BibTeX, RIS and CSV for Zotero, Mendeley, EndNote and LaTeX workflows. FileMind organizes and indexes the loose files; Zotero remains the citation manager you write with.

Zotero and Mendeley are reference managers: you import papers into their library and they can rename the imported copies from their metadata. FileMind works on the folders you already have, without an import step, and adds what reference managers do not: an evidence trail and confidence score on every proposed name, journaled one-click rollback, semantic and hybrid full-text search, and question answering with citations from your own papers. It is a one-time purchase and complements, rather than replaces, a reference manager.

Yes. FileMind scores the text quality of every page and routes pages with little or garbled text to a bundled OCR engine (PaddleOCR) that runs locally. Scanned papers become searchable and are marked Partial OCR or OCR only so you can filter on them.

Windows 10 or later today. A macOS build is planned and every license already covers both platforms, so there is nothing further to buy when it ships. Linux support is planned.

A typical modern laptop. 8 GB of RAM is the minimum and 16 GB is recommended if you run a local AI model. A GPU is optional; it speeds up local AI, OCR and embeddings but is not required. The installer is about 880 MB, the app takes about 2.5 GB installed, and the optional built-in AI model adds about 2.5 GB.

FileMind is a one-time purchase, not a subscription. The public beta is free with no PDF limit. On the paid release the free trial covers up to 100 PDFs, Pro is $39 for a personal license on 2 machines, and Lab is $139 for up to 5 seats. Both include lifetime updates and a 30-day refund.

A built-in local model (Qwen3-4B served by llama.cpp, set up in one click), any model running in Ollama or LM Studio, cloud APIs from OpenAI, Anthropic (Claude) and Google (Gemini), or any OpenAI-compatible endpoint such as Azure OpenAI or vLLM. Local models are free and keep everything on your machine; cloud providers are optional and FileMind can cap their monthly spend.

No. FileMind at filemind.app is a research-paper organizer first released in 2026. It is unrelated to the discontinued FileMind file manager from 2013 that some third-party download sites still list. Only download FileMind from filemind.app or its GitHub releases page.