Skip to content

Answers from your documents, with sources

A knowledge assistant that searches your SOPs, contracts and help articles first, then answers with a link to the exact source.

RAG knowledge assistants

Starting price
$6,000
Typical timeline
5 to 6
weeks to launch

Works with

Op

When knowledge is hard to find

Answers live in 500 files

The answer exists somewhere in Drive, Notion or an old PDF. Finding it takes 20 minutes.

The same questions every day

New staff ask senior staff the same questions, pulling them away from their own work.

General AI makes things up

Public chatbots do not know your policies and will confidently guess when they are unsure.

How retrieval-augmented generation works

It looks up your documents first, then answers. That is what stops it from guessing.

Question

What notice does a tenant need to give before ending a lease?

Your document library

Leave policy.pdf

Lease template.docx

Onboarding SOP

Maintenance guide

Price list 2026

Support FAQ

Answer

Tenants must give 60 days' written notice before the lease end date. Notice is sent to the property manager by email or post.

[1] Lease template, clause 14.2[2] Maintenance guide, p. 3

What we build for you

Document library

We connect your Google Drive, Notion, PDFs, help centre and website as the assistant's source material.

Cited answers

Every answer links to the document and section it came from, so anyone can check it.

Internal help desk

Staff ask questions about policies, processes and products and get answers in seconds.

Customer support

Customers get accurate answers from your help articles, with handoff to your team when needed.

Access control

People only get answers from the documents they are allowed to see.

Automatic updates

When a document changes, the assistant's knowledge updates on the next sync.

Example: an internal assistant for a property manager

Before

Staff searched 300+ Drive documents or asked the operations manager about lease terms and maintenance procedures.

After

Staff ask the assistant in plain words and get the answer with a link to the exact clause or procedure.

An illustrative example of a typical project.

Safety and accuracy, built in

AI is only useful if you can trust it. These rules are part of every build, not optional extras.

Grounded answers

The assistant only answers from retrieved passages. If nothing relevant is found, it says so.

Sources on every answer

Each answer links to its source, so people can check rather than trust blindly.

Permission-aware

Search respects your access rules. Confidential files stay confidential.

Measured accuracy

We test against real questions before launch and share the results with you.

Tools we connect

All integrations
  • OpOpenAI
  • Anthropic
  • Google Drive
  • Notion
  • Supabase
  • LangChain
  • Intercom
  • n8n

What you receive

  • A connected and cleaned document library
  • The assistant, in Slack-style chat, on your site or as an internal web app
  • Source citations on every answer
  • Access rules matched to your team structure
  • An accuracy test set and report

Timeline

  1. CollectWeek 1

    We gather the documents, remove outdated ones and agree who can see what.

  2. IndexWeeks 2 to 3

    We split, label and index the documents so the right passages are found.

  3. EvaluateWeek 4

    We test against a set of real questions with known answers and measure accuracy.

  4. LaunchWeeks 5 to 6

    We roll it out to a pilot group first, then to everyone.

RAG knowledge assistants

From $6,000

for an assistant over one document library with citations and access rules. AI usage is billed at cost.

Questions we hear often

What does RAG mean?

Retrieval-augmented generation. The AI first retrieves relevant passages from your documents, then writes an answer based only on them.

Which documents can it use?

Google Docs, Sheets, PDFs, Word files, Notion pages, help centre articles and web pages. Scanned documents can be converted to text first.

Is our data used to train AI models?

No. We use business API terms that exclude your data from model training, and your index is stored in your own account.

How accurate is it?

We measure it on your own test questions before launch. Because every answer shows its source, errors are easy to spot and fix.

Find the first workflow worth automating

A 20-minute call is enough. We tell you plainly what we would automate first, and what it would cost.

What happens on the call

  1. 1We look at how the work happens today
  2. 2We name the first workflow worth automating
  3. 3You get a fixed written quote. No obligation

Free, 20 minutes, on Google Meet.