Service
A custom AI chatbot that answers from your company knowledge and shows its sources
Staff or customers ask a question in plain language. The AI assistant finds the relevant passages in your manuals, policies and product sheets, writes an answer from them and cites the source. When nothing in your documents supports an answer, it says so.
- Built for
- Answers that must come from your own material
- Method
- Retrieval from your documents, with citations
- Staff role
- Own the content, review unanswered questions
- Price
- Request a quote
For companies whose answers are buried in documents
A general AI tool answers from what it learned in training. A RAG chatbot answers from your documents: it retrieves the relevant passages first, then writes from them. RAG stands for retrieval-augmented generation. It is built when the answer has to match your manual, your policy or your product sheet.
- Companies with large manuals or policy sets
- Technical distributors with many product sheets
- Schools
- Law and accounting firms
- BPO training teams
- Customer support groups
How it works
Agree
Sources and access rules
We agree on which documents are included and which groups may see which collections.
Index
Documents prepared and indexed
Documents are cleaned, split into passages and indexed in a private store.
Retrieve
A question retrieves passages
Each question pulls the most relevant passages the person asking is allowed to see.
Answer
An answer with citations
The AI model answers from those passages and cites them, so the reader can open the source.
Refuse
A clear reply for gaps
A question with no supporting passage gets a "not found in the documents" reply and a contact to ask.
Improve
Usage reviewed
Usage and unanswered questions are reviewed with the content owner to improve the sources.
From a question to a cited answer
- Question asked (Slack, Teams or web), Automated
- Access group checked (Permissions), Automated
- Relevant passages retrieved (Private index), Automated
- Answer written with citations (AI model), Automated
- No passage: "not found" reply (Refusal rule), Automated
- Content owner reviews gaps (Review), Your staff
Custom GPT, knowledge base or a custom build
Three routes lead to an assistant that knows your material. The right one depends on how many documents you have, who may see them and how much checking the answers need.
Custom GPT
A custom GPT or project
A custom GPT, or the equivalent in Claude or Gemini, holds a small set of files and instructions. It suits one team and a modest document set, and you can set it up yourself.
Knowledge base
An AI-powered knowledge base
Help desk and wiki tools now include AI search over their own articles. It suits content that already lives in that tool.
Custom build
A custom chatbot on your sources
A build over your drives and systems, with access by group, citations, a refusal rule and a test report. It suits large or restricted document sets.
How answers are kept to company information and kept current
The assistant is instructed to answer only from retrieved passages, and it is tested against sample questions with known correct answers before anyone relies on it. Keeping it current is a content task: when a document changes in the source folder, the index is refreshed, and the content owner retires what is outdated.
- A test report shows how the assistant answered your sample questions
- Access controls are tested with restricted documents before launch
- Unanswered questions are listed for the content owner, as a guide to missing documents
What we need from you and what we handle
We need from you
- The documents and who may see them
- Sample questions with correct answers, for testing
- A content owner who keeps the sources current
- A decision on where staff or customers will ask: Slack, Teams or the web
We handle
- Reviewing provider data terms with you before documents are processed
- Preparing and indexing the documents
- The chat interface, citations and the refusal rule
- Access by group, tested with restricted documents
- The test report and a walkthrough for the content owner
What a first project includes and what it leaves out
Included
- One knowledge collection
- One chat interface
- Source citations
- Access by group
- A test report against your sample questions
Left out
- Digitising paper archives
- Legal review of the documents
- Training a model from scratch
Limits we state plainly
- Answers can still be wrong when the source is unclear or outdated. Citations let users check, and high-stakes answers need a person.
- Confidential documents are sent to an AI provider for processing. Provider data terms, retention and data location are reviewed with you, and a self-hosted model is an option at higher cost.
- Access controls must be tested so restricted documents do not leak into answers.
- It is not a source of legal, medical or financial advice.
Tools and channels involved
The Claude, OpenAI or Gemini API as the AI model, a vector database as the private index, Google Drive or SharePoint as sources, a chat interface in Slack, Teams or on the web, and n8n to keep the index refreshed. Simple reference data can also be read from a sheet, as described on the Google Sheets integration page.
What it costs
Cost depends on the size of the document set and how access is divided.
Custom AI knowledge assistant
Request a quote
Quoted per project after a short review of your channels, volumes and systems.
Request a quoteWhat moves the price
- Number and condition of the documents
- Number of collections and access groups
- Chat interfaces needed
- AI provider usage, or a self-hosted model
- Testing depth
- Support level
Questions we get asked
How is a custom chatbot kept to company information?
It retrieves passages from your documents first and is instructed to answer only from them. Each answer cites its source, and the setup is tested against questions with known answers.
What happens when the answer is not in the documents?
The assistant replies that it did not find the answer in the documents and points to a person or channel to ask. The question is logged for the content owner.
How is the knowledge kept current?
The index is refreshed when documents change in the source folders. A content owner on your side decides what is added, updated or retired.
Is this different from a custom GPT or custom ChatGPT?
A custom GPT suits a small file set for one team. A custom build adds access by group, a larger document set, citations you can audit and a chat inside the tools your staff already use.
Can customers use it, or only staff?
Either. A customer-facing version uses only documents approved for the public and hands over to staff for anything else.
Where are our documents processed?
By the AI provider you approve, under its business data terms, which are reviewed with you first. A self-hosted model is an option at higher cost.
Next step
Tell us which documents your team keeps searching through.
Describe the document set, who may see it and the questions asked most. We will say which route fits and what a first collection would cover.