Ask once. Read the answer, with its sources attached.
Rudani searches your own documents and writes the answer. Every sentence carries the passage it came from, and anything the documents cannot support, it declines to say.
Watch an answer being assembled.
Pick a question. Rudani retrieves, re-reads, then writes, threading each claim back to the document underneath it.
Your documents are indexed where they sit.
Rudani reads from the systems you already run. Nothing is copied into an index you do not control.
Retrieval is ranked, then re-read.
Candidate passages are read in full before a single word is written, so a ranking mistake does not become an answer mistake.
Every sentence is written against a passage.
A claim with no passage behind it does not reach the page. The model is allowed to say that the documents do not answer the question.
A citation points at the paragraph, not the file.
Open any number in the answer and you land on the sentence it came from.
It runs where your documents already are.
In your cloud or ours
Self-hosted, in your VPC, or operated by us. Your keys, your logs, your retention policy.
Indexes what you already have
SharePoint, S3, Drive, Confluence, Postgres, and flat scans out of the archive.
Answers in the language asked
The languages your archive is written in, with the passage quoted in its original.
The rest of the system
Point it at your own archive.
We index a sample of your documents and hand you the questions it can already answer.