Ask your documents. Keep them yours.
A private RAG assistant: it indexes what you upload, retrieves the right passage, and answers with citations you can open. OCR and vision read scans and photographs, Whisper transcribes audio and video, and your own database answers in plain English. Runs on your own LLM key — OpenAI, Claude, Gemini — or self-hosted, so documents never leave your server. Every account is isolated; sharing is an invitation, never a default.
Private by construction
Every account is isolated. Your documents are indexed for you alone and sent only to the model provider you chose, under your own key.
Answers you can open
Every figure and sentence is traced to the passage it came from. Click a citation and read the source, with the matching words highlighted.
Upload once, shared across the workspace
One copy of the document, queried by everyone you share it with. Offer a colleague or your team your knowledge bank, they accept, and they can ask across it without you sending anything and without a second copy on their side — one upload, one index, counted once against your storage, and every citation names whose document it came from. Either side can switch it off at any time.
Ask your database in plain English
Connect PostgreSQL, MySQL, Supabase or a SQLite file, pick the tables and columns it may see, and ask. The SQL it writes is shown to you and runs read-only.
Charts computed, not guessed
Ask for a chart and get one built by SQL over your own rows, not numbers a model typed out. Pie, bar, line, scatter — with the query on the card so you can check it.
Files, not just chat
Excel with live formulas, CSV, Word, and designed PDF reports with your logo and brand colour. Thirty-eight formats in all, from a spreadsheet to a Python script.
Reads scans and photographs
Vision and OCR for documents with no text layer: a scanned statement becomes a queryable table, and a value you ask about is ringed on the page it came from.
Web search when you want it
Switch it on per conversation. Sources are read and cited, and what was read is kept so asking again costs nothing.
Share with a friend, or a whole team
Offer a colleague the use of your knowledge bank or your model key. They see the offer, accept it, and can switch it off any time. Nothing moves until they say yes.
Once they accept, they can simply ask — and get answers out of your documents without you sending them anything and without a second copy on their side. One upload, one index, one storage bill. Between two friends that is a convenience; across a workspace or a department it is the difference between one shared library and the same contract uploaded nine times.
- Ask across shared documents — no re-upload, no duplicate storage
- Every answer says which document, and whose it is
- Offer, accept, switch off, withdraw — either side, any time
- What others spend on your key is shown to you, per provider
One knowledge bank, a team working in it
Upload once and everyone you invited asks against the same library. Charts, spreadsheets and PDFs the assistant makes stay attached to the conversation, and the usage page shows what each account spent.
- Per-account usage and cost, down to the answer
- Two-factor sign-in and session protection for every member
- Admin console for plans, alerts and audit
What a private RAG service is, and what makes this one private
RAG — retrieval-augmented generation — means the assistant looks up the answer in your documents before it writes, and cites the passage it used. A private RAG keeps those documents, the index built from them, and the questions asked of them inside one account, on infrastructure you control.
Is PrivateRAG a private RAG SaaS or a self-hosted app?
Both. Use it as a hosted private RAG SaaS on your own model key, or run the same private RAG app on your own server when documents cannot leave the building. One codebase, one behaviour.
What does a private RAG app need to read?
PDFs, Word, Excel and PowerPoint; scanned pages through OCR; photographs and diagrams through a vision model; audio and video through Whisper; and a read-only connection to your database, answered in plain English with the SQL shown.
How is a private RAG service different from asking a chatbot?
Every figure and sentence is traced to a source you can open. A number that cannot be traced is flagged, not guessed. Charts are computed by SQL over your rows, not typed out by the model.
- Indexed for your account alone — no shared corpus
- Your own OpenAI, Claude or Gemini key — tokens are never resold
- Citations you can open, with the matching words highlighted
- Hosted, or self-hosted on request
Who builds this
PrivateRAG is made by Zobitas, a software company founded by David Wagih, its sole founder, CEO and CTO, in United Arab Emirates. We build one thing: a private assistant that reads the documents you give it and answers out of them, with the source attached.
Zobitas is a software company. It has no connection to any construction, insulation or waterproofing business with a similar name.
- Your documents are indexed for your account alone
- Model calls run on your own provider key — we do not resell tokens
- Self-hosted on request, for teams that cannot send documents anywhere