pdf-forge

Windows PDF toolkit — merge, split, reorder, rotate, delete, compress, and extract pages from PDFs with a fast, friendly desktop UI. Optional local-AI adds OCR for scanned PDFs and page/document summarization using tiny local models. Offline & privacy-first — your documents never leave your machine.

Overview

pdf-forge is a lightweight Windows desktop app (Windows 10/11 first) for the everyday PDF chores that normally push people toward sketchy online converters:

• Merge several PDFs into one, in any order.
• Split a PDF into multiple files (by page ranges, every N pages, or one-per-page).
• Reorder / rotate / delete pages via a visual page-thumbnail board (drag & drop).
• Extract selected pages to a new document or to images (PNG/JPG).
• Compress to shrink file size (downsample images, strip unused objects).
• Metadata — view/edit title, author, subject, keywords.
• Optional local-AI: OCR a scanned/image-only PDF into a searchable text layer, and summarize a page or whole document — all via a local model endpoint. Fully optional and off by default.

Everything works 100% offline. There is no account, no cloud upload, no telemetry.

Motivation

Basic PDF editing on Windows is a pain: the built-in tools can't merge or reorder, desktop suites are heavy/paid, and free online tools mean uploading private documents to a stranger's server. pdf-forge keeps all of it local, fast, and free — a single small app that does the 90% of PDF tasks people actually need.

Use cases

• Combine scanned receipts / invoices into one file for expenses.
• Pull a few pages out of a big report to share.
• Rotate pages that were scanned sideways.
• Reorder a document that came out of the scanner backwards.
• Shrink a huge image-heavy PDF so it fits an email attachment limit.
• Make a scanned contract searchable with local OCR (no cloud).
• Get a quick local-AI summary of a long PDF before reading it.

How to use (Windows-first quickstart)

Status: early scaffolding. Build steps below are the target workflow; see milestones.

1. Download the latest portable pdf-forge-win-x64.zip from Releases (or the MSIX installer).
2. Unzip and run PdfForge.exe — no install required for the portable build.
3. Drag one or more PDFs onto the window (or File → Open).
4. Use the page board to reorder / rotate / delete pages, or the toolbar for Merge / Split / Extract / Compress.
5. Click Save As to write the result. Originals are never modified in place.

Build from source (developers)

git clone https://github.com/rwrife/pdf-forge.git
cd pdf-forge
dotnet build -c Release
dotnet run --project src/PdfForge.App

Build Windows packaging artifacts

# Portable self-contained folder
dotnet publish src/PdfForge.App/PdfForge.App.csproj -c Release -r win-x64 --self-contained true

# MSIX package (Windows + MSBuild with Desktop Bridge tooling)
msbuild packaging/PdfForge.Package/PdfForge.Package.wapproj /restore /p:Configuration=Release /p:Platform=x64 /p:AppxPackageSigningEnabled=false

Tag pushes matching v* run .github/workflows/release.yml, which publishes:

• pdf-forge-win-x64.zip (portable self-contained build)
• pdf-forge-win-x64.msix (MSIX installer)

Example workflow

Merge two PDFs and extract pages 3–5 of the result:

1. Drag a.pdf then b.pdf onto the page board — pages append in order.
2. Select pages 3, 4, 5 on the board (Ctrl+click).
3. Extract → Selected pages → New document.
4. Save As pages-3-5.pdf.

OCR a scanned PDF (local-AI on):

1. Open the scanned PDF.
2. Tools → Make searchable (OCR) — pdf-forge renders each page and runs the local OCR model, then writes an invisible text layer.
3. Save. The output is now selectable/searchable in any reader.

Local-AI integration

pdf-forge runs fully without AI. When you opt in, it talks to a local, OpenAI-compatible endpoint (e.g. Ollama or llama.cpp server) on http://localhost:11434 — nothing is sent to the cloud.

• OCR / scanned-PDF reading: a tiny local vision model (MiniCPM-V class) or a native OCR fallback (Windows.Media.Ocr) turns page images into text.
• Summarization: a small local text model (Llama 3.2 3B / Qwen2.5 / Phi-3-mini class) summarizes a page or document; only extracted text is sent to the local endpoint.
• A reachability probe checks the endpoint on demand; if it's unavailable, AI features gray out gracefully and the rest of the app is unaffected.
• All AI features are off by default and configurable (endpoint URL, model name) in Settings.

Current status / milestones

• M0 — Repo scaffold, solution layout, CI on windows-latest
• M1 — Core PDF engine: load, render thumbnails, save (initial implementation)
• M2 — Merge / split / extract / reorder / rotate / delete
• M3 — Compression + metadata editor
• M4 — Page board UI (drag & drop, multi-select) + drag-drop file open
• M5 — Optional local-AI: OCR searchable layer + summarization
• M6 — Packaging: portable win-x64 zip + MSIX, first release

See PLAN.md for scope, architecture, and non-goals.

---

Part of the auto-tool-lab Windows utilities series. Privacy-first, local-only, no telemetry.