GClaude Indexer

Name: GClaude Indexer — Download free. Local document classification
Availability: InStock
Rating: 5 (1 reviews)
Author: alexccastilho

GClaude Indexer — Free Download. Local document classification

GClaude Indexer converts folders of scanned PDFs, Word files, Excel sheets, PowerPoint presentations, images, emails and plain text into structured Markdown indexes without uploading content to external servers. The application runs OCR on pages lacking a text layer, splits oversized PDFs into readable segments, extracts text from every page and classifies the material into discrete items with type, date, author and summary. Four separate Markdown files are generated: a complete item index, a chronological timeline, a review report detailing gaps and failures, and project instructions ready for pasting into an AI workspace. Original documents remain untouched. The tool operates on Windows 10 and 11 with an installer that configures Python, Tesseract, Ghostscript and Ollama automatically.

★★★★★

5.0(1 ratings)

Download GClaude Indexer (Official links)

File size: 7.77 MB

The latest version of GClaude Indexer is: 1.3.0

Operating system: Windows

Languages: English

Developer: alexccastilho

Price: $0.00 USD (Open Source (GPL-3.0))

Folder scanning with recursive document detection. The engine walks through the chosen source directory and identifies supported file types across all nested subfolders. It builds an internal inventory of every document before processing begins, which allows the progress bar to show accurate percentages and prevents duplicate work when the same file appears in multiple locations through shortcuts or hard links. The scan phase also records file sizes, modification dates and MIME types so that later steps can decide which conversion route to apply.
OCR execution on scanned pages without text layers. Each PDF page and image is examined for an embedded text layer. Pages that already contain extractable text are passed directly to the reading stage, while pages that contain only rasterized content are routed through Tesseract OCR. This selective approach reduces processing time significantly because OCR is not run on pages that already have machine-readable text. The OCR output is written to a temporary working directory and never alters the source document.
Oversized PDF slicing into readable chunks. Documents that exceed a configured page threshold are divided into smaller logical sections before classification. The slicer respects paragraph boundaries and heading positions where possible, avoiding cuts that would split a sentence or separate a caption from its image. Each slice receives a stable identifier that links back to the original page range, so the final index can point to an exact location inside a large file.
Text extraction from Word, Excel, PowerPoint and email formats. The extractor handles legacy binary Office formats as well as modern Open XML files. For Excel workbooks, each worksheet is processed separately and cell coordinates are preserved in the output. For emails, headers such as sender, recipient, subject and date are parsed into structured fields. Extracted text is normalized to UTF-8 and stripped of control characters before being passed to the classification engine.
Content classification into typed items. The core function assigns each extracted segment a type, a date when one is detectable, an author and a short summary. The rules engine uses deterministic pattern matching for dates, names and document types. The local Ollama engine uses a language model to infer metadata when explicit patterns are absent. The output schema is the same regardless of which engine produced the classification, so results from different engines can be compared directly.
Markdown index generation. The index file lists every classified item with its type, date, author, summary and source document reference. Each entry includes a link to the original file and the page number where the item was found. The index is sorted alphabetically by type and then by date, making it suitable for direct use in documentation systems or static site generators.
Timeline construction ordered by date. The timeline report extracts all items that contain a recognizable date and orders them chronologically. Items without dates are listed separately in an undated appendix. The timeline helps identify document sequences, correspondence chains and project phases that span multiple source files. It can be regenerated independently of the other reports after reclassification.
Review report with gap and failure listing. After processing, the tool produces a report that lists every file that could not be read, every page where OCR confidence fell below a threshold, and every item where a required field could not be determined. Failures are grouped by cause, with the original filename and page number for each entry. This report serves as a quality control checklist before the output is used for further work.
Project instruction generation for AI workspaces. The fourth output file contains a set of instructions written in Markdown that describes the document collection, its structure, the classification metadata and how to use the generated reports. These instructions are formatted for direct pasting into an AI project context, allowing a language model to answer questions about the indexed documents without accessing the original files. The instruction file includes the index and timeline as references.
Local classification with Ollama engine. The default classification route runs an open model through Ollama on the local machine. The Ollama host is fixed to 127.0.0.1 in the code and is not read from configuration, ensuring that no document text, page content or filename is transmitted over the network. The default model is qwen3.5:4b, approximately 3.2 GB to download, and the system automatically falls back to the rules engine when memory is insufficient.
Deterministic rules engine without external dependencies. This engine requires no model download, no GPU and no network access. It uses regular expressions and pattern libraries to identify document types, extract dates in multiple formats, detect author names from signatures and email headers, and generate summaries from the first substantive sentence of each section. The rules engine runs on any Windows machine that can execute Python and serves as a baseline for comparing model-based results.
Claude Code integration for network-based classification. The fourth classification option routes work through Claude Code and is the only engine that uses the network. This engine sends extracted text to the Claude API for classification and returns structured metadata. It is suitable when higher-quality classification is needed for non-sensitive collections and the user has API access. The other three engines never leave the machine.
Local web interface in three languages. The application serves a browser-based interface at 127.0.0.1:8000 with English, Brazilian Portuguese and Spanish translations. Language selection is available from a header dropdown and the default is detected from the Windows display language. The interface includes four color themes that persist in browser storage between sessions.
Live resource monitoring with CPU, RAM and GPU graphs. During execution, the interface displays real-time graphs of CPU usage, RAM consumption and GPU utilization. GPU temperature, power draw and clock speed are shown when available without administrator rights. An optional CPU sensor shortcut adds CPU temperature and power readings through an elevated helper process that runs separately from the main application.
Pause and resume control per processing step. Each row in the execution screen has its own pause button and progress bar with a time estimate. Pausing a step stops the current operation cleanly at the next safe boundary and preserves all completed work. Resuming continues from where the step stopped without reprocessing completed items. The run button for each step can be triggered independently, allowing the user to rerun only the classification stage after changing engines.
Quality comparison report across engines and models. The tool records the declared confidence and field completion rate for each classification engine and model against the same collection. A comparison view shows side-by-side metrics for rules-based, Ollama and Claude Code results. The quality score measures confidence and completeness, not factual correctness, and is intended as a diagnostic aid rather than a validation measure.
Shared project catalog via Google Drive. The Projects screen includes a shared catalog option that points to a folder inside Google Drive. When configured, every computer signed into that Drive account can open, edit and delete the same projects. Without this setting, the project list remains local to the machine even though project files themselves sync through Drive. Projects located on another computer's local disk are listed and marked as out of reach.
Disk space recovery for intermediate files. After the four reports are generated and the user has reviewed them, a button on the Result screen deletes intermediate OCR'd PDFs and text slices. The database, the four report files and the logs are never removed by this action. The feature is designed for collections where disk space is limited and the intermediate working files are no longer needed.
Installer with automatic dependency setup. The first-run installer checks for Python 3.12 and installs the official 3.12.10 for the current user if it is missing, without requiring administrator rights or affecting any other Python installation on the machine. Tesseract and Ghostscript are installed automatically for OCR support when the user agrees. Ollama and its default model are offered as optional components. The installer also creates a desktop shortcut and an optional elevated shortcut for CPU sensor readings.
Uninstaller with selective removal. The uninstall script distinguishes between components owned by the application, such as the virtual environment and shortcuts, and separately installed programs like Tesseract, Ghostscript, Ollama and Python. Each item is offered individually with the default answer being no. Project folders and generated reports are never deleted, and the script lists their locations with sizes for the user to decide on separately.

Development of GClaude Indexer began as a response to the need for processing scanned document collections without moving confidential material to external services. The project was initiated by a developer working with large archives of paper-origin files that required classification and indexing before being fed into AI-assisted workflows. The initial version focused on OCR and text extraction only, with classification added later as the value of typed metadata became apparent. The rules engine was the first classification method, followed by Ollama integration and finally the Claude Code option. The codebase is written in Python with PowerShell scripts handling Windows-specific tasks such as hardware detection, registry access and Performance Counters monitoring. The project is distributed under the GNU GPL-3.0 license and remains actively maintained on GitHub.

Alternatives to GClaude Indexer:

GClaude Indexer

Alternatives to GClaude Indexer:

Glyfo

SnipFor

Relic