Your documents — answered on the phone.
Upload PDFs, price lists, manuals. The AI uses them as knowledge for every call, email and chat — with source citation, always current. Included in your voiceOne plan from £29/month.
Why a knowledge base for your AI?
Because ChatGPT doesn’t know your prices. And the printed price list at reception is from 2024. A RAG knowledge base solves both.
From your real documents
Price lists, T&Cs, product datasheets, FAQ lists, internal phone directory — what you upload, the AI knows. Updates available instantly.
With source citation
The AI says ‘according to price list 2026, page 4’. You can always verify where an answer came from. Hallucinations practically impossible.
Always current
Upload a new price list version? AI uses it immediately. Old version archived. Conflicts flagged.
Stays with you
Your documents stay on voiceOne servers in UK/EU. No upload to OpenAI, no training use, no data export.
Features of the RAG knowledge base
A full retrieval pipeline without DIY effort.
PDF, Word, Excel, HTML, Markdown
Drag & drop and the AI understands. Tables, lists, headings, embedded images with OCR.
- Up to 500 MB per file
- Auto-OCR on scans
- Tables parsed structurally
Your website as a source
Specify URLs, we crawl and keep fresh. Ideal for FAQ pages, help centres, blog content.
- Daily cron
- Sitemap-based
- Token for restricted areas
API to your ERP / PIM
Price list, stock, lead time direct from your system. Live query, no re-index needed.
- JSON/REST/SQL connectors
- Caching with TTL
- Auth tokens stored securely
One source, many languages
An English source serves German, French, Spanish callers automatically. Translation with source link.
- 13 languages out of the box
- Industry vocabulary
- Per tenant isolated
What the AI may say
Stop-words, banned topics, escalation triggers. Industry-specific defaults (legal: no legal advice; medical: no diagnosis).
- Compliance templates
- Audit log per answer
- Sandbox mode for testing
AI learns with you
When the AI couldn’t answer a question 5x, it auto-suggests a new FAQ entry.
- Weekly gap analysis
- Top unanswered queries
- Draft new entries
voiceOne RAG vs. ChatGPT Custom GPT vs. DIY (LangChain)
Three ways to ground AI on your documents. What you really need.
| Feature | voiceOne RAG | ChatGPT Custom GPT | DIY (e.g. LangChain) |
|---|---|---|---|
| Entry price | £29/mo (in plan) | $20/mo (Plus) | Servers + dev from £1,500/mo |
| Phone integration | Built-in | No | DIY |
| Source citation in call | Yes | Sometimes | DIY |
| PDF/Word upload | Direct UI | In UI | DIY |
| Web crawler | Included | Not out of box | DIY |
| API to your systems | REST/SQL ready | Per Action API | Free DIY |
| Server in EU/UK | UK/Germany | US (sub-processor) | Self-hostable |
| Data not used for training | Contractually | Enterprise | Self-hosted |
| Auto-refresh | Included | Manual | DIY |
| Compliance templates | Industry setups | No | No |
As of 05/2026 — all prices per the providers’ public pricing pages. We refresh this table quarterly.
How our customers use it
Three real configurations.
Chartered accountants Manchester
200 pages of client FAQ, T&Cs, fee schedule as RAG. The AI answers standard queries on deadlines, document submission, mandate status — with direct reference to the relevant rule.
Outdoor retailer Bristol
Product PIM with 4,800 SKUs as API source. AI knows lead time, stock, warranty live. Phone orders pre-qualified, cross-sells suggested.
Commercial chambers London
Internal client FAQ as RAG. But: NEVER legal advice on the phone. AI only gives process guidance and availability. Stop-words ‘claim’, ‘file’ trigger escalation.
Included in your voiceOne plan
Talk to a RAG specialist
30 minutes with our RAG specialist. We analyse which sources are worth it and which compliance guards are needed.
Frequently asked questions
Retrieval-Augmented Generation. Unlike ChatGPT, which only knows its training data, a RAG system first searches YOUR documents on every question and then answers based on those concrete hits. Prevents hallucinations and guarantees current answers.
They are stored on voiceOne servers in UK/EU, converted to semantic vectors, and used only to answer your own calls/emails. NOT for AI training, NOT for other customers, NOT shared with OpenAI/Google.
(1) Every answer is annotated with the exact source quote. (2) If the AI doesn’t find a sufficient source, it honestly says ‘I don’t know’ or escalates. (3) Stop-words block certain topics entirely.
Up to 500 MB per file. Standard plan covers 500 documents or about 200,000 pages. Larger needs go into a Custom plan.
Yes. Sandbox mode: upload test documents and ask 50 questions. We show answer and source citation, you rate. The result decides if the knowledge base goes live.
Yes. The knowledge base also works for chatbot, email AI and internal wiki. Full impact comes with phone integration.
Standard plan: 500 documents, 100,000 answers/month. Above: £0.005 per answer, or Custom plan from £199/month for 10x volume.
Knowledge base ready in 30 minutes.
Upload, test in sandbox, go live. 14-day free trial.