Answers from your own documentsRAG systems and AI knowledge bases
A private AI assistant that answers from your own documents, policies and data, shows where each answer came from, and keeps your information yours.
WHAT IT DOESLOG
“What’s our refund policy?”
→ Answered, with the page it came from
New starter question
→ Answered from the handbook
Clause across 200 contracts
→ Found in seconds
Product spec lookup
→ Exact figure, linked to the sheet
“The answer is in a PDF somewhere. Or a shared drive, an old email, or one colleague’s head, and they’re on holiday.”
SOUND FAMILIAR?
What is a RAG system?
A RAG system (retrieval-augmented generation) lets an AI assistant answer questions from your own documents instead of from general internet knowledge. When someone asks a question, the system finds the most relevant passages in your handbooks, policies, contracts or product information, then has an AI model write the answer from those passages and show where it came from. The result is a private AI knowledge base: accurate, up to date when your documents change, and honest when the answer isn’t there.
Tamkoz builds RAG systems and AI knowledge bases for businesses and individuals. We gather your sources, set who can see what, build an assistant that quotes its sources, and test it against the questions your team and customers really ask. Your documents stay under your control, on services that don’t train on your data.
THE JOBS IT TAKES ON
RAG & AI KNOWLEDGE BASES
“What’s our refund policy?”
→ Answered, with the page it came from
New starter question
→ Answered from the handbook
Clause across 200 contracts
→ Found in seconds
Product spec lookup
→ Exact figure, linked to the sheet
Not in the documents
→ Says so, instead of guessing
What a private AI knowledge base solves
Most businesses already have the answers. They’re just scattered, buried or stuck in one person’s head.
The answer is in a PDF somewhere, but nobody knows which
→ Ask in plain language and get the answer with the page it came from.
New starters asking the same questions for months
→ An assistant trained on your handbook and processes answers them, day one.
Customers getting different answers from different people
→ Every answer drawn from the same approved policy, every time.
Staff pasting company documents into public chatbots
→ A private assistant on your own documents, with your access rules.
Hours searching contracts, specs or case files
→ Search across hundreds of documents at once, by meaning, not just keywords.
A generic chatbot that makes things up
→ Answers only from your sources, and a clear “I don’t know” when they don’t cover it.
Ask it about your own documents
A letting agent’s staff assistant, answering from the company handbook and tenancy templates.
Harbour Lettings
STAFF ASSISTANT · YOUR DOCUMENTS ONLY
What just happened
A new team member asked three questions. Here’s how the assistant answered, and what it refused to guess:
SAMPLE CONVERSATION · FICTIONAL LETTING AGENCY
Who uses RAG systems?
Anyone whose work depends on finding the right answer in a lot of documents.
Property and letting agents
Tenancy terms, compliance rules and property files answered in seconds.
Law firms and accountants
Precedents, guidance notes and client files searched by meaning, with sources to check.
Clinics and care providers
Policies, procedures and training material at staff fingertips, kept private.
Manufacturers and distributors
Product specs, safety sheets and manuals answered for staff and trade customers.
Schools, trainers and course creators
Course material turned into an assistant learners can question.
Support teams
Help articles and past answers behind every reply, so agents and chatbots stay consistent.
Researchers and professionals
Your own papers, notes and files, searchable and questionable without a public chatbot.
A GOOD FIT IF…
- Your knowledge lives in PDFs, drives and inboxes
- Staff ask the same internal questions
- Customers need accurate answers about products or policies
- Your data can’t go into a public chatbot
What we build
A source map
Which documents and systems feed the assistant, who owns each one, and who may see what.
A clean, indexed library
Your documents prepared, split into passages and indexed, with updates picked up as files change.
An assistant that cites its sources
Answers written from your passages, with links back to the document and section.
Access rules
Permissions that follow your existing groups, so people only see what they’re allowed to.
A test set of real questions
The questions people really ask, with checked answers, re-run whenever the system changes.
Where people already work
On your website, in Slack or Microsoft Teams, inside your help desk or as a private web app.
How we build a RAG system
Good answers start with good sources, so that’s where we start.
- ORDER 001START
1×Gather the sources
We list where your knowledge lives (handbooks, PDFs, shared drives, help articles, product sheets) and agree what’s in, what’s out and who may see what.
- ORDER 002BUILD
1×Prepare and index
Documents are cleaned up, split into sensible passages and indexed so the right passage can be found in a fraction of a second.
- ORDER 003CHECK
1×Build the assistant
We connect the index to a model that answers only from your passages, quotes its sources, and says when the answer isn’t there.
- ORDER 004GO LIVE
1×Test with real questions
We collect the questions your team and customers really ask, check every answer, and tune until the results hold up.
Where your knowledge can come from
We connect to the places your documents already live, so nobody has to copy files into a new tool.
- FILES AND DRIVES
- Google DriveSharePointOneDriveDropboxPDF and Word files
- WIKIS AND HELP CENTRES
- NotionConfluenceZendesk GuideIntercom articlesYour website
- WHERE PEOPLE ASK
- Website chatSlackMicrosoft TeamsHelp deskPrivate web app
- AI MODELS
- OpenAIAnthropic (Claude)Google (Gemini)Open models you host
- SEARCH AND STORAGE
- Vector databasesPostgreSQL with pgvectorYour own cloud account
If you have strict data rules, tell us early. We can keep everything inside your own cloud account and use providers that don’t train on your data.
How much does a RAG system cost?
Building the system is a one-off job we quote in writing. Running it costs a small monthly amount you pay directly: indexing your documents, storing the index, and the AI model answering each question.
WHAT AFFECTS THE PRICE
How many documents, and how messy
A tidy folder of PDFs is quick. Scans, spreadsheets and years of mixed files need more preparation.
How many sources
One shared drive is simple. Several systems, each connected and kept in sync, take longer.
Access rules
One audience is simplest. Different permissions for different teams or customers add work.
Where it lives
A web chat is quickest. Slack, Teams or help-desk integration adds a little.
Privacy requirements
Running inside your own cloud account or with a self-hosted model costs more than a managed service.
FROM THE PROVIDERS THEMSELVES
- Indexing text is cheap. OpenAI’s text-embedding-3-small model costs $0.02 per million tokens, so indexing a 500-page handbook (roughly 300,000 tokens) costs less than one cent. Source: OpenAI.
- Answering questions uses a language model, priced per million tokens. Anthropic’s smallest current Claude model costs $0.10 per million input tokens for prompts up to 100,000 tokens. Source: Anthropic.
CHECKED 10 OCTOBER 2026. PROVIDERS CHANGE THEIR PRICES, SO CHECK BEFORE YOU BUY.
YOUR QUOTE · FIXED AND IN WRITING
We quote the build in writing, in your currency, and estimate monthly running costs for your document volume.
How long does it take?
A first version on one set of documents can often be answering questions within a few weeks. More sources are added in stages.
The longest step is usually tidying the documents themselves. We’ll show you which files will cause wrong answers, and why.
TYPICAL TIMELINE
YOUR QUOTE HAS THE REAL DATES
01 SOURCE REVIEWONE CALL
We look at your documents together and agree what goes in.
02 QUOTEA FEW DAYS
Scope, privacy approach, price and dates in writing.
03 FIRST VERSIONTWO TO FOUR WEEKS
Indexed, built and tested against your real questions.
04 ROLL-OUTSTAGE BY STAGE
More sources, more users and new places to ask, once the first is trusted.
RAG and AI knowledge base questions
Something else? Ask in the chat. A person answers.
What is a RAG system?
RAG stands for retrieval-augmented generation. When someone asks a question, the system first retrieves the most relevant passages from your own documents, then gives them to an AI model to write the answer from. The answer is grounded in your information, can quote where it came from, and stays up to date when your documents change.
What is a RAG chatbot?
A chatbot that answers from your documents using RAG. Staff or customers ask in plain language and get an answer from your handbook, policies or product information, usually with a link to the page it came from, rather than a guess from the model’s general knowledge.
Is RAG better than training or fine-tuning a model on our data?
For answering questions from documents, usually yes. RAG is quicker to set up, cheaper to keep current (you update a document, not a model), and can show its sources. Fine-tuning is better for teaching a model a style or a narrow task. Many good systems use RAG alone.
Is ChatGPT a RAG system?
Not by default. ChatGPT answers mainly from what its model learned in training. Some features, like uploading files or searching the web, work in a RAG-like way. A RAG system built for your business searches your own documents every time, under your access rules.
Will our documents be used to train someone else’s AI?
We build on business plans and settings where the AI provider doesn’t train on your data, and we tell you which provider holds what. If your documents must not leave your own infrastructure, we can run the system on your cloud account or with an open model you host.
Can different people see different documents?
Yes. Access rules can follow your existing permissions, so HR files stay with HR and customers only ever see what’s published for them.
What happens when the answer isn’t in our documents?
It says so. A well-built RAG assistant is told to answer only from the passages it finds, and to point people to a person when they don’t cover the question. Those gaps are logged, which is a useful list of what your documentation is missing.
What kinds of files can it use?
Most of them: PDFs, Word and Excel files, slides, web pages and help-centre articles, and content in Google Drive, SharePoint, Notion or Confluence. Scanned documents can be read with text recognition first.
Can customers use it, or is it only for staff?
Both. An internal assistant helps your team find answers fast. A customer-facing one, on your website or help centre, answers product and policy questions from your published information, and hands over to a person when needed.
Can individuals use a RAG system?
Yes. Researchers, consultants, writers and professionals with large personal libraries use one to search and question their own notes, papers and files without uploading them to a public chatbot.
GOES WELL WITH
- AI customer serviceAn AI customer service agent that resolves routine questions across chat, email and your help desk, follows your policies to the letter and hands real problems to your team.
- AI assistants & chatbotsA chat assistant on your website, WhatsApp, Instagram or Facebook that answers customers in seconds, in your words, day and night.
- AI integrationWe connect models like ChatGPT, Claude and Gemini to your website, CRM, app or internal systems, so AI works on your data, inside your workflow, under your control.
SOURCES · CHECKED 10 OCTOBER 2026
Put your knowledge to work.
Tell us where your documents live and who needs answers from them. We’ll come back with a plan and a written quote.