docsearch

Index local documents into a ChromaDB vector store with OpenAI-compatible embeddings.

Updated Mar 24, 2026

Search Tags:#embeddings #semantic-search #local-docs #information-retrieval #chromadb #document-indexing

Ready to run

npx skills add https://github.com/thevgergroup/mad-skills --skill docsearch

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: docsearch
Source: https://github.com/thevgergroup/mad-skills/tree/main/skills/docsearch
Command: npx skills add https://github.com/thevgergroup/mad-skills --skill docsearch

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires chromadb, requests, docx, pptx, pypdfium2, openpyxl, and includes scripts (resource) components.

What problem does it solve?

Index local documents into a ChromaDB-based vector store to enable fast, semantic search across large collections of PDFs, Word, slides, spreadsheets, and text files.

Core Features & Use Cases

Ingest a directory tree and generate chunked embeddings for efficient retrieval.
Perform semantic searches with optional metadata filtering and hierarchical depth-based scoping.
Incrementally re-index as files change, preserving existing results while updating only new content.

Quick Start

Run a simple quick start by indexing your documents with the built-in script and then performing a search query.

Frequently Asked Questions about docsearch

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I perform semantic search across local PDF and Word documents?▼

You can perform semantic search by indexing your local PDFs, Word files, and spreadsheets into a ChromaDB vector store using chunked embeddings. This allows you to query content based on meaning rather than exact keywords.

Can I re-index only the new or changed files in my document directory?▼

Yes, you can incrementally re-index changed files. This preserves existing vector store results while updating only the new or modified content, ensuring your searchable index stays current without redundant processing.

What document formats are supported for indexing into a vector database?▼

Supported document formats include PDF, Word, PowerPoint slides, spreadsheets, and text files. These formats are parsed and chunked to generate embeddings for reliable information retrieval in the vector store.

Do I need an OpenAI-compatible endpoint to generate embeddings for local docs?▼

Yes, an OpenAI-compatible embeddings endpoint is required. The tool uses this environment-configured backend to generate chunked embeddings for your local documents before storing them in the ChromaDB vector database.

How do I filter semantic search results by file metadata and directory depth?▼

You can filter semantic search results using optional metadata filtering and hierarchical depth-based scoping. This relies on per-file metadata captured during indexing to narrow down retrieval to specific directories or file attributes.