Corporate Knowledge Assistant (RAG)
A corporate knowledge assistant answers questions from your organisation's own documents (procedures, contracts, product catalogues, internal policy) and cites its source for every reply; it does not place orders, open records, or make decisions, it only knows and answers accurately. Deployed on web, WhatsApp, or Slack/Teams, it runs on infrastructure you control and filters access by user role.
In scope
- Document ingestion and vectorization (PDF, Word, Excel, web pages, FAQs)
- Embedding and semantic search infrastructure setup
- Role-based access filtering (department, permission level)
- Source-cited answer generation (which document, which section)
- Channel integration: web, WhatsApp, or Slack/Teams (one or combined)
- Confidence-based handoff to a human when the answer is uncertain
- Team training and documentation handover
Out of scope
- LLM usage costs and server/hosting (billed at cost)
- Action-taking agent capabilities — placing orders, updating records — that's AI Agent Development
- Improving the quality of source documents themselves (used as-is)
- Ongoing content updates (covered by the monthly maintenance plan, not the one-off project)
Evidence
- P2For a healthcare education provider, 21 medical textbooks and the organisation's full procedure set were indexed into a vector database; 913,531 past conversations and 275 web pages were included, with every answer grounded via classification and semantic search.
- P1A packaging manufacturer's multilingual WhatsApp assistant answers Turkish, English, and Arabic questions around the clock using the company's own product procedures; after-hours lead loss dropped to zero.
Sectors
Pricing
| Scope | Duration | Range |
|---|---|---|
| Corporate Knowledge Assistant (RAG) | 3-8 weeks | 2,500 - 8,000 € |
| Monthly maintenance and monitoring | — | 500 - 2,000 € |
- Prices exclude VAT.
- LLM usage costs and server/hosting are not included; billed at cost.
Frequently Asked Questions
Does our data leave our systems?
No. The assistant runs on infrastructure you choose, your own servers or a cloud region you select, and your documents are never used to train a third-party model. Only the minimum context needed to generate an answer is sent to the LLM provider; the raw document store stays with you.
Can it run on our own servers?
Yes, self-hosting is a standard option. n8n, the vector database (e.g. Supabase/pgvector), and the orchestration layer can all run on your infrastructure; only the LLM API call leaves your network, scoped to anonymised context.
What happens if it gives a wrong answer?
Every answer is shown with its source, so users can see which document it came from and verify it. When confidence is low, the system does not guess, it hands the question to a human instead; the threshold is tuned per project.
What file formats can it read?
PDF, Word, Excel, PowerPoint, web pages, and plain text are supported by default. Scanned (image) PDFs are processed with OCR, and structured data sources (databases, APIs) are integrated separately depending on project scope.
How many documents can it handle?
Scale depends more on total page volume and query load than document count; we have worked on projects from a few hundred to several thousand pages, including one reference case indexing 21 books and 275 web pages in a single system. We size the architecture to your volume.
Does it connect to our existing systems?
Yes. It integrates with CRMs (HubSpot, Salesforce), help desks, internal document stores (SharePoint, Google Drive), and ticketing systems; a note can be logged automatically to the relevant record once an answer is generated.
How is this different from an AI agent?
A knowledge assistant only knows and answers with sources cited; it does not place orders, update records, or make decisions. If you need a system that takes action and chooses between multiple tools on its own, that's AI Agent Development.
Also read
Last updated: 7 August 2026
Talk About Your Knowledge Assistant