What is RAG (Retrieval-Augmented Generation)?

RAG is an AI technique where the model retrieves relevant passages from your knowledge base before generating a response. This grounds answers in your real business data rather than general AI training, eliminating hallucinations.

What content sources can I add to the 1CS knowledge base?

You can connect any website URL (the crawler indexes pages automatically), upload PDF or document files, or add FAQ entries manually. All sources are searchable by the AI together.

How is semantic vector search different from keyword search?

Vector search understands meaning and context, not just exact word matches. It finds relevant answers even when the customer phrases their question very differently from how you wrote the content.

Do AI answers include source references?

Yes. Every AI response includes the source it was retrieved from — the URL, document name, or FAQ entry — so agents and customers can check the original content.

What vector search technology does 1CS use?

The knowledge base is powered by OpenAI embeddings combined with HNSW (Hierarchical Navigable Small World) vector indexing — the same approach used by leading AI search systems.

How often does the knowledge base update when my website changes?

You can trigger a re-sync manually at any time, or configure automatic re-crawling on a schedule. The indexing pipeline processes updates in minutes.

Knowledge Base RAG

AI answers from your real data. Not guesses.

Your AI answers from your actual website and business content using vector search technology used by leading AI systems — not hallucinations.

Start free → See pricing

Knowledge Base

4 documents · last synced 2m ago

Indexed

🌐

Main website

18 pages

📄

Article: Return policy

Article

❓

FAQ

24 items

🌐

Products & pricing page

synced 2m ago

Cheapest package — how much?

AI Answer

The Starter plan is free — includes 100 credits. No credit card required.

Source: Products & pricing page

The most common failure of AI-powered support tools is hallucination — the AI generates a confident-sounding answer that is simply wrong. This happens when the AI relies on its general training data rather than your specific business information. 1CS solves this with Retrieval-Augmented Generation (RAG): before generating any response, the AI retrieves the most relevant passages from your knowledge base and uses only that retrieved content to construct its answer. Every response is grounded in your real data and cites the source it came from.

Key capabilities

✓ Connect your website, articles, or FAQs — AI reads and indexes them automatically
✓ Semantic vector search finds relevant answers, not just keyword matches
✓ Answers include source references you can trace back
✓ Powered by OpenAI embeddings + vector search (HNSW)

How Knowledge Base RAG works

Four steps from setup to first resolution.

01

Connect your content sources

Paste any website URL — the crawler automatically fetches and indexes every linked page. You can also upload PDF or Word documents, or write FAQ entries directly in the 1CS interface. Multiple sources of different types can coexist in the same knowledge base.
02

Automatic vectorisation and indexing

1CS processes your content using OpenAI text-embedding models, converting each passage into a high-dimensional numerical vector that encodes its meaning. These vectors are stored in an HNSW (Hierarchical Navigable Small World) index — the same vector search architecture used by leading AI systems — enabling millisecond retrieval across thousands of documents.
03

Semantic retrieval at query time

When a customer asks a question, 1CS converts the question into a vector and finds the passages in your knowledge base whose vectors are most similar — meaning most semantically related — to the question. This works even when the customer's phrasing bears no literal resemblance to how you wrote the content.
04

Grounded answer generation with source citation

The AI generates a response using only the retrieved passages as context. The final answer includes a citation: the URL, document name, or FAQ entry that the answer came from. Agents and customers can follow the citation to read the original source.

RAG vs fine-tuning: why RAG wins for business support

Fine-tuning trains the AI model itself on your data — an expensive, time-consuming process that must be repeated every time your content changes. RAG separates the retrieval step from the generation step: the AI model stays unchanged, while your knowledge base is updated in minutes. For a business with frequently changing pricing, policies, or product catalogues, RAG is the only practical approach to keeping AI answers current.

Semantic vector search vs keyword search

Traditional keyword search matches words literally. A customer asking "how do I get my money back?" would not match an article titled "Refund Policy" unless both use identical words. Semantic vector search understands meaning: it recognises that "get my money back" is semantically equivalent to "request a refund" and retrieves the correct article regardless of the exact phrasing. This dramatically improves answer accuracy for natural language questions.

Source references and trust

Every AI answer generated by 1CS includes a reference to the source passage: the URL that was crawled, the document name that was uploaded, or the FAQ entry that was matched. This transparency serves two purposes: it lets your support team verify that the AI is answering correctly, and it allows customers who want more detail to go directly to the source document rather than relying on the AI summary alone.

Keeping your knowledge base current

Website content changes constantly — new pricing, updated policies, seasonal promotions. 1CS lets you trigger a manual re-sync of any URL at any time, or configure automatic re-crawling on a recurring schedule. The indexing pipeline processes updates within minutes, ensuring the AI is always answering from your latest published content rather than an outdated snapshot.

Who uses this

Teams that get the most value from Knowledge Base RAG:

→ Product teams indexing documentation so AI answers technical questions accurately
→ E-commerce stores letting AI answer shipping, return, and policy questions from live web pages
→ Service businesses uploading pricing PDFs so the AI quotes correctly on every channel
→ Support teams reducing ticket volume by surfacing existing help articles automatically

Set up Knowledge Base RAG

Step-by-step guides from the 1CS Help Center:

Add content to the knowledge base

Upload documents, crawl your website, or write FAQs

Configure AI Agent Brain

Connect your knowledge base to the AI agent

Frequently asked questions

Everything you need to know about Knowledge Base RAG in 1CS.

What is RAG (Retrieval-Augmented Generation)?: RAG is an AI technique where the model retrieves relevant passages from your knowledge base before generating a response. This grounds answers in your real business data rather than general AI training, eliminating hallucinations.
What content sources can I add to the 1CS knowledge base?: You can connect any website URL (the crawler indexes pages automatically), upload PDF or document files, or add FAQ entries manually. All sources are searchable by the AI together.
How is semantic vector search different from keyword search?: Vector search understands meaning and context, not just exact word matches. It finds relevant answers even when the customer phrases their question very differently from how you wrote the content.
Do AI answers include source references?: Yes. Every AI response includes the source it was retrieved from — the URL, document name, or FAQ entry — so agents and customers can check the original content.
What vector search technology does 1CS use?: The knowledge base is powered by OpenAI embeddings combined with HNSW (Hierarchical Navigable Small World) vector indexing — the same approach used by leading AI search systems.
How often does the knowledge base update when my website changes?: You can trigger a re-sync manually at any time, or configure automatic re-crawling on a schedule. The indexing pipeline processes updates in minutes.