Talk to PDF

A secure document Q&A workspace that lets teams upload PDFs, ask natural language questions, and get clear answers backed by citations from the original document. Built for contracts, reports, manuals, research papers, and internal knowledge documents.

GitHub

Overview

What it does

Upload PDFs and turn them into searchable Q&A workspaces.
Ask questions in plain English instead of manually scanning long documents.
Get concise answers grounded in the uploaded file.
Check citations and source passages behind every answer.
Keep questions organized by document, project, or topic.
Review token usage, latency, retrieval details, and answer evidence when needed.

Turn PDFs into searchable workspaces

Create a separate workspace for each uploaded PDF.
Keep contracts, reports, papers, and manuals organized.
Open, rename, or delete document projects as needed.
Track useful document details such as file name, size, and upload time.

Talk to PDF document parsing pipeline showing local and cloud PDF extraction options.

Extract text and structure from complex PDFs

Supports both self-hosted and cloud-based document extraction.
Uses local parsing when privacy or deployment control matters.
Can use Azure Document Intelligence for stronger layout-aware extraction.
Prepares the document for accurate search, retrieval, and citation-backed answers.

Talk to PDF question thread management screen.

Keep document questions organized

Create separate question threads for different topics or review tasks.
Return to previous questions without losing context.
Avoid messy one-long-chat document exploration.
Fine-tune retrieval settings when deeper investigation is needed.

Talk to PDF answer screen showing a concise response from the uploaded PDF.

Ask questions and get PDF-grounded answers

Ask natural language questions about the uploaded document.
Receive short, readable answers instead of long raw excerpts.
See how many sources were used for the response.
Designed to avoid confident answers when the document does not provide enough evidence.

Talk to PDF details screen showing tokens, latency, retrieval, and rewrite information.

Inspect how each answer was produced

Shows token usage for prompt and generated answer.
Breaks down latency across query rewriting, retrieval, and generation.
Displays retrieval strategy and selected settings.
Helps developers and teams monitor cost, speed, and answer quality.

Talk to PDF source citation screen showing retrieved chunks and reranking scores.

Check citations and source evidence

Shows the source passages used to generate each answer.
Lets users verify the answer against the original PDF.
Displays retrieval or reranking scores for source evidence.
Makes the system easier to trust, debug, and improve.