FileMind

Getting Started

Get FileMind up and running in under five minutes. This guide covers installation, first-run setup, and your first library scan.

System Requirements

MinimumRecommended
OSWindows 10 / macOS 12Windows 11 / macOS 14
RAM8 GB16 GB (for larger AI models)
Disk2 GB free5 GB+ (AI model + embeddings)
GPUNot requiredAny CUDA/Metal GPU speeds up AI

Installation

Windows

  1. Download FileMind-setup.exe from the download page
  2. Run the installer and follow the prompts
  3. FileMind appears in your Start menu and system tray

macOS

  1. Download FileMind.dmg from the download page
  2. Open the DMG and drag FileMind to Applications
  3. Launch from Applications — macOS may ask you to confirm since it's not from the App Store

First Launch

On first launch, FileMind runs the Setup Wizard automatically. The wizard installs Ollama (the local AI runtime), downloads an AI model matched to your hardware, and verifies everything works with a test query. The whole process takes about 2-5 minutes depending on your internet speed.

Your First Scan

  1. Click Add Library Folder in the toolbar (or drag a folder onto the window)
  2. Select the folder containing your PDFs — FileMind scans recursively through subfolders
  3. Watch the scan progress: FileMind extracts text, detects metadata, and generates embeddings
  4. When the scan completes, rename proposals appear in the Rename Queue

FileMind processes files in the background. You can keep working while it scans — the UI updates in real time as papers are indexed.

What Happens During a Scan

For each PDF, FileMind:

  1. Extracts text — native extraction for digital PDFs, OCR (via Tesseract) for scanned documents
  2. Scores text quality — determines if OCR is needed based on character quality
  3. Detects identifiers — looks for DOI, arXiv ID, ISBN, and other identifiers
  4. Fetches metadata — queries CrossRef, arXiv API, and other sources using detected identifiers
  5. Parses heuristically — extracts title, authors, year from document structure when APIs don't help
  6. Falls back to LLM — uses the local AI model only when deterministic methods fail
  7. Generates embeddings — creates vector representations for semantic search and related-paper discovery
  8. Creates a rename proposal — builds a filename from extracted metadata with a confidence score

Next Steps