Customer supportFeatured

Document Q&A Agent

Answers from your documents, and says nothing when they do not answer

Version 3Published Sep 2, 2026
Watch it work▶ Try it on your own figures — free

5 of 8 questions need attention.

8 questions checked. 5 flagged. The largest gap is 29 days.

Start with How long does invoice approval take. HR has the most flagged.

Whether an answer is correct is for its owner; the list shows which questions are open, unsourced or unconfirmed.

5
5/8
Flagged
5
8
Flagged
5
Longest (days)
29
Share flagged
63%

Why rows were flagged

5
  • Unanswered for over 7 days33%
  • Answered without a source33%
  • Not confirmed by an owner33%

Value flagged by Topic

  • HR2
  • Finance1
  • Support1
  • Legal1

Items to act on, worst first

ItemTopicValueDays waitingWhy
How long does invoice approval takeFinance129Unanswered for over 7 days
What are the remote work rulesHR128Unanswered for over 7 days
When is a refund allowedSupport10Answered without a source, Not confirmed by an owner
How long is data keptLegal10Answered without a source
What is the notice periodHR10Not confirmed by an owner

What to do this week

  • Resolve How long does invoice approval take first: it is the most urgent on the list.
  • Fix the data or the status on every flagged row, then re-run.
  • Re-run this check every time the list changes, before it reaches a decision.

Do

  • Fix the highest-risk item first, then run the check again on fresh data.

Don't

  • Don't act on one row alone: open the source record first.

How to improve

  • Add more detail per row (owner, date, source) so the next report can trace each item.
  • Add a second period of data so the next report can show whether things are improving.

Demo

Try it now — free

Type your figures (or fill in samples) and get the answer in a minute.

Start from a template

Make it decision-ready

Also include

No card · no API key · your figures are not stored

Overview

Asking questions of your own documents is one of the largest demands in this whole market, and almost every implementation shares one silent failure: when the documents do not contain the answer, the model answers anyway — from what it knows about the world, in exactly the confident voice it uses when quoting your own contract. The reader cannot tell which is which, which makes it worse than a search box, because a search box that finds nothing says so.

This answers only from what it retrieved, and quotes it. Every claim carries the passage and the document it came from, so an answer can be checked in seconds rather than trusted.

When your documents do not answer the question, it says so and stops. It also says when they answer it in two different places with two different answers, rather than silently picking one — because in a company that has been running for a few years, contradictory documents are not an edge case, they are the normal state, and choosing between them is not a machine's job.

Benefits

Answers people act on

Because each one is traceable to a line in a real document.

The same answer every time

Whoever asks, and whoever is on holiday.

You find your own gaps

The questions nothing answers are the documents you are missing.

What it does

  • Quotes the passage behind every claim

    An answer you can check in seconds rather than one you have to trust.

  • Stops when the documents are silent

    The whole product. Every other implementation answers from general knowledge in the same confident voice.

  • Reports contradictions instead of choosing

    Two documents with two answers is the normal state of any company a few years old. Picking one is not a machine's job.

  • Answers from current versions only

    A superseded policy cannot be quoted back at you, because it is never retrieved.

  • Respects who is asking

    It can only reach documents the person asking is allowed to open. That is enforced outside the agent.

  • Says what would answer it

    When nothing covers the question, it names the document that would.

What this changes

What this changes about your week

It takes over
Interrupting the person who knows, or reading forty pages to find one line.
Done today by
Everyone, several times a day.
How often
Continuously.
Typical manual effort
Minutes each time, from two people instead of one.

Why Customer support in particular. Every other implementation answers from general knowledge when your documents are silent, in the same confident voice it uses to quote your contract — and the reader cannot tell the difference.

  • Runs unattended

    Started by you or by an event, and it finishes on its own. Nothing waits for someone to be at a desk.

  • The same standard every time

    The two-hundredth item is held to the bar the first one was. Consistency is the part people cannot sustain.

  • It cannot act on its own

    Document Q&A Agent has no path to sending, spending or committing. That limit is why its output is safe to act on.

How it runs

This agent runs server-side through the PROMIVO runtime. Each run is logged step by step and every tool call is permission-checked before it executes.

  1. 1InputValidated against the agent input schema
  2. 2Modelclaude-opus-5 with automatic fallback
  3. 3Toolssearch_knowledge
  4. 4PermissionsRuns READ actions automatically; anything higher waits for approval
  5. 5OutputStructured result matching the output schema
  6. 6LoggingTokens, duration and every step recorded

Requirements

Integrations

No integrations required.

Before you start

  • A knowledge base holding the documents
  • The question
  • Requires the Starter plan or higher

Use cases

Internal helpdesk

Policy and process questions, answered from the policy.

Onboarding

New starters ask the documents instead of interrupting someone.

Contract and policy lookup

What does our agreement actually say about this.

Pricing

Starter

Billed monthly through your PROMIVO subscription. Cancel at any time.

Runs consume your plan allowance for agent executions and tokens. See plan limits

FAQ

What happens when the answer is not in our documents?

It says so and names what would answer it. It will not fall back on general knowledge — that is the failure this exists to avoid, and it is invisible to the person reading.

Can it see documents a person should not?

No. Retrieval is scoped to what the person asking may open, and that is enforced in the retrieval layer rather than by instructing the agent. An instruction can be talked around; a permission check cannot.

What about outdated documents?

Superseded versions are never retrieved, so a withdrawn policy cannot be quoted back at you. The old text is kept, so an answer given last month can still be explained.

Reviews

Reviews open once customers have run this agent.

Need this agent to work differently?

Tell us what to change and our team will scope a customised version for your business.

Customize this agent