Law and professional services Document Q&A Retrieval with a guard

Chat with your case files, and it refuses to guess.

Ask a folder of contracts, emails and reports a plain question. Every answer shows the file and the exact page, with the clause marked on it.

One question can read across a contract, an email thread and a survey at once, and cite each one.

1

A question comes in

Plain English. No keywords to guess.

2

Pages ranked by meaning

Every page scored against the question.

3

A cutoff before the model

Too low, and the server answers on its own.

4

Answer, with the page

The cited clause, marked where it sits.

Not in the files? It says so

No page clears the cutoff, so the model is never asked. No invented citation.

The problem

Finding the one clause in three hundred pages eats the afternoon.

An associate opens the contract, then the email thread, then the survey. They hunt for the clause, the reply that answers it, and the number that ties it together.

A plain chatbot will happily answer, and sometimes it invents the citation. Lawyers have been sanctioned for filing exactly that.

This build answers from the files only, cites the exact page, and stops itself when the answer is not there. You get the speed without betting your filing on a made-up quote.

  • Page-exact citations. Every answer opens the page it came from, with the clause marked.
  • No invented citations. A cutoff on the server decides, not the model's goodwill.
  • Reads across files. One answer can stitch a clause, an email and a report together.
Try it live

Ask the case files a question.

Shown here on a sample dispute. The same engine reads your own contracts, emails and reports. Simulated case, real retrieval and citation guard.

basaltflow.ai/case-files Live

This runs the same engine I would build for you. Ask the settlement question and watch it refuse.

Want this running on your own documents?

Tell me what piles up: contracts, email threads, reports. I will reply with the fastest honest way to make it answerable.