InRag: the most advanced end-to-end RAG solution
InRag, short for Intranet RAG, is the most advanced and complete end-to-end solution for implementing AI retrieval-augmented generation (RAG), ready to deploy. It turns your organization’s documents into a private knowledge assistant. It covers ingestion, search, answers, administration and the apps your staff use, and it runs on your own server, inside your corporate network (your intranet), where your documents already live.
Answers from your documents, with the proof attached
Building RAG yourself means assembling and maintaining many separate components, then making them work together. InRag delivers all of it as one product. Staff ask questions in plain English, about policies, agreements, minutes, letters, financial statements and forms, and InRag finds the right passages and writes an answer from them alone.
Each fact comes with an exact quote and a link that opens the original file at the cited page. If your documents don’t contain the answer, InRag says so.
From folder to verified answer
Four steps, all managed from one admin console.
-
Connect
Point InRag at your folders of PDF, Word, PowerPoint, Excel and other files, including scanned pages, and group them into collections such as HR, Legal or Finance.
-
Build the knowledge base
InRag reads and indexes every document on your own server, and keeps the knowledge base current as files are added, changed or removed.
-
Ask
Staff ask questions in plain English, including follow-ups, in the web chat or the Windows app.
-
Get a verified answer
InRag answers from your documents alone, quotes each fact word for word with a link to the page, and protects personal identifiers from cloud models.
Everything a production deployment needs
Ingestion
- OCR only where it’s needed, with OCR text flagged in answers
- Collection-wide letterhead and footer removal
- Folder monitoring, resync and schedules across time zones
- A report for every run: files processed and any failures
Search and conversation
- Multi-turn chat that shows users what was actually searched
- Whole-document questions such as “summarize the letter”
- A short list to choose from when several documents fit
- People’s other names: “Robert J. Avery”, “Bob” and “RJAvery” are one person
Trust and privacy
- Exact quotes, verified word for word
- An honest “not found” instead of a guess
- Privacy tokenization for identifiers and pay amounts
- Optional general-knowledge answers, always labelled “Not from your documents”
- Citations that open the original file
Administration
- Department-level access to collections
- Live settings that apply within seconds, with no restart
- Per-collection ingestion settings
- A log of every question with its sources and timing
- Notices to users when a collection is updated
Built once. Always current.
Processing documents is the expensive step, so InRag does it once per document. After that, every question is answered from the knowledge base, and updates touch only the documents that changed.
-
Ingest once
Each document is read and indexed a single time and stored in the knowledge base on your server.
-
Ask without limits
Any number of users can ask any number of questions. Each one searches the prepared knowledge base, with no reprocessing.
-
Update on schedule
Daily, weekly or monthly schedules in any time zone, plus a folder monitor that checks for changes every few minutes.
-
Only what changed
Unchanged files are skipped, a changed file is updated on its own, and a deleted file is removed from the knowledge base.
Why it matters
- Fast answers. Questions never wait for documents to be processed.
- Low running cost. Updates take minutes, not a full rebuild, and the server isn’t tied up re-reading files that haven’t changed.
- Always up to date. Answers reflect the latest version of each document, and removed documents stop appearing in answers.
- Nothing to babysit. Users are notified when a collection changes, and quiet runs that find nothing new send no notices.
Only 4 documents were read again. Everything else in the knowledge base was left untouched, and the collection’s users were told what changed.
Your server, your network, your choice of model
InRag installs on a single machine in your office network: Windows Server or a Mac. Use a local model and nothing leaves the machine, or a leading cloud model with privacy tokenization on.
- Web chatA browser chat with sign-in, chat history and a document viewer.
- InRag Chat for WindowsA native desktop chat app.
- Web and Windows adminSources, runs, schedules, settings and users.
- Self-service accountsEmail sign-up codes and password reset in every app.
Questions buyers ask
Does any of our data leave our network?
With a local model, no: documents, questions and answers stay on the InRag server. With a cloud model, only the question and the relevant passages are sent, with Social Security numbers, account and card numbers, contact details, dates of birth, ID numbers and pay amounts replaced by placeholders first. Document processing always runs on your server.
How do we know an answer is right?
Every fact must carry an exact quote from your documents. InRag checks each quote against your documents, removes any it can’t find, and flags answers with no verifiable quote. Each citation opens the original document at the cited page.
What happens when the documents don’t have the answer?
When your documents don’t contain a relevant passage, InRag replies that it found nothing instead of guessing. General-knowledge answers are available only if a user turns them on, and they are clearly marked.
Who can see which documents?
Documents are organized into collections. A collection is open to every signed-in user or limited to one department, and other users can neither see nor search it.
Do we have to re-process everything when documents change?
No. Documents are ingested once. On each scheduled run or folder check, InRag checks every file for changes: unchanged files are skipped, changed files are updated, new files are added, and deleted files are removed.
Where does InRag read documents from?
From folders on or attached to the InRag server. Folders are monitored for added, changed and deleted files, and runs can be scheduled daily, weekly or monthly. Connectors to systems such as SharePoint are not included today; we can discuss your sources during a proof of concept.
Which languages are supported?
InRag is built and measured for English-language documents today. Other languages are supported on request: tell us which languages your documents use when you request a demo or contact us.