Chat with Your Documents

A conversational AI interface for document Q&A and knowledge extraction. Ask questions over your files and over read-only databases in the same chat.

Chat with Your Documents combines two things in one chat surface: ask questions over your uploaded documents with inline citations, and ask questions over your live databases that auto-render as Plotly charts.

What you can do

Ask questions across documents

Upload PDFs, DOCX, and XLSX files. The system reads them, splits them into chunks, and indexes them. Ask anything in any language and get answers with inline [1] [2] citation chips you can click to inspect the source chunk.

Query live databases

Connect a read-only SQL source (Postgres, MySQL, SQL Server, Snowflake, BigQuery, and others). Ask questions, the system writes the SQL for you, runs it, and renders the result as a Plotly chart inline.

Mix documents and SQL in one turn

Some questions need both. The agent picks Auto, RAG (documents), or SQL per turn, and shows tool tags (RAG, SQL · , RAG + SQL, RAG · HyDE) so you always see which path was taken.

Start here

Welcome

The page you land on the first time you open the product.

Knowledge Bases

Create a knowledge base and add documents.

SQL Data Sources

Connect a read-only database to chat with.

Chat Canvas

Start a new conversation.

Pages in this system

Welcome

Lands here on first sign-in. Picks your first project and forwards you to its Overview. The empty-roster state shows a Create new project CTA.

Project Overview

Per-project home with knowledge-base count, document count, SQL-source count, and recent chats.

Knowledge Bases

Document collections with primary language and chunking settings. Three-tab detail (Documents / Scope summary / Settings), plus create and delete.

Documents

PDF, DOCX, and XLSX uploads with a status pipeline (pending, parsing, chunking, embedding, ready or failed). Per-doc tags, scope info, and a chunk inspector with inline edit.

SQL Data Sources

Read-only database connections — your credentials are stored encrypted. Schema / Semantic terms / History / Connection tabs. Bilingual semantic glossary so you can teach the agent your terminology.

Chats

Conversation history with Active, Archived, and Deleted views. Sessions bind the knowledge bases and SQL sources you chose.

Chat Canvas

The single-session view. Composer with drag-drop attachments, force-tool dropdown, scope-override chips, and streaming responses.

Project Settings

Project name, deletion (with typed-keyword confirmation), permissions, sharing.

Share Links

Visitor-facing chat surface. Chatbot or snapshot mode, optional password gate, configurable scope, admin read-back of visitor sessions.

Tool transparency tags

Every assistant message carries one or more tags so you know how the answer was produced.

Tag	Meaning
RAG	Pulled from your uploaded documents.
SQL ·	Wrote and ran SQL against the named data source.
RAG + SQL	Combined both in the same answer.
RAG · 2 sub-queries	Decomposed your question into multiple retrieval steps.
RAG · HyDE	Generated a hypothetical answer first to guide retrieval.

The tags stay in English on Arabic pages (they’re technical labels).

Common questions

What file types can I upload?

PDF, DOCX, and XLSX. There’s a per-file size limit shown on the upload screen, plus an overall storage cap per project. Other formats are not supported today.

Can I share a chat with someone outside Wittify?

Yes, via Share Links. You can publish either a chatbot mode (visitors can ask new questions, scoped to whatever you allow) or a snapshot mode (visitors can read a frozen conversation but can’t continue it).

My SQL source is read-only, but is it really safe?

Yes — read-only access is enforced at multiple levels. You connect with read-only credentials, and any data-modifying SQL is blocked. Even so, only connect to databases where read-only access is acceptable.

The answer cited a document but I can't find the quote in my file.

Click the citation chip in the answer. The chunk inspector opens with the exact source text the answer was built from. If you’ve updated the file since, re-upload it so the index reflects the new content.

Why did the agent pick SQL when I expected it to use my documents?

You can override per turn. Click the tool dropdown next to the composer and pick RAG (documents), SQL, or Auto. Or use the scope chips to narrow which knowledge bases or sources are in play for that turn.

Can the agent answer questions in mixed Arabic and English?

Yes. The embedding model supports 100+ languages and the agent mirrors your question’s language back at you in the answer. Mixed-language documents are supported, and the system never strips Arabic diacritics or normalises bidi characters.