TableSnap (Windows Starter Project)

Capture a table from anywhere on screen, reconstruct its rows, columns, and merged cells, correct the result, then copy it into Word or Excel or export an .xlsx file. Recognition is connected through ITableRecognizer; the starter project includes an OpenAI Responses API implementation. The demo table works without an API key.

Run in VS Code

1. Install the .NET 10 SDK for Windows. The runtime alone is not enough to build the project.

2. Install VS Code and Microsoft's C# extension. C# Dev Kit is optional.

3. Open the TableSnap folder in VS Code.

4. Run these commands in the VS Code terminal:

   dotnet --version
   dotnet restore
   dotnet run

You can also press F5; the project includes .vscode/tasks.json and .vscode/launch.json.

Enable image recognition

Create a file named .env in the project root (next to TableSnap.csproj) and add your key:

OPENAI_API_KEY=your-api-key
TABLESNAP_MODEL=gpt-4o

Restart TableSnap after editing .env: exit from the system tray, then run dotnet run again. The .env file is ignored by Git. An existing system environment variable takes priority over the value in .env.

Alternatively, set the API key in the PowerShell terminal that runs the app:

$env:OPENAI_API_KEY = "your-api-key"
dotnet run

Do not put the key in source files or commit it to Git. TABLESNAP_MODEL can override the model name; the default is gpt-4o. Captured images are sent to the configured API. The request uses image input and JSON Schema structured output; see the OpenAI image input guide and structured output guide.

Use the app

• Press Ctrl+Shift+2 or click Capture Region to select a visible table. Press Esc to cancel.
• Click Open Image to recognize an existing screenshot.
• Click Demo Table to test editing and merged-cell export without an API key.
• Edit the recognized cell text directly in the table preview.
• Click Copy Table to place HTML table, TSV text, and CSV data on the clipboard. Paste into Word or Excel.
• Click Export XLSX to create an editable workbook. It preserves the grid, row and column merges, bold text, and alignment, and adds uniform basic borders.
• Closing the main window leaves the app in the system tray. Right-click the tray icon to exit.

Recognition corrects common grid-coordinate mistakes such as zero spans, one-based positions, and underreported row or column counts. The status message flags adjusted results for careful review. Conflicting or oversized cells are rejected with a specific cell number instead of silently discarding data.

Run the offline structure tests without an API key:

dotnet run --project Tests/TableDocumentTests.csproj

Current scope

This starter project preserves table structure and selected basic styles. It does not reconstruct exact fonts, background colors, column widths, complex borders, or numeric cell types. Exported cells are stored as text so that values such as 0012, percentages, and thousands separators are not silently changed by Excel. Review recognition results before using them. Test capture coordinates on mixed-DPI multi-monitor setups.

The project has no external NuGet dependencies. Models/TableDocument.cs defines the internal table structure, Recognition/ITableRecognizer.cs allows other recognition engines, and Export/XlsxExporter.cs writes an Office Open XML workbook directly.