Knowledge Base

The knowledge base lets agents search your documentation, FAQs, and internal content during conversations. When a caller asks a question, the agent retrieves the most relevant passages and uses them to answer — instead of guessing or escalating immediately.

This is called Retrieval Augmented Generation (RAG): the agent is grounded in your content rather than relying solely on the LLM's training data.

How it works

1. You add content to the knowledge base — by uploading files, pasting text, or connecting a URL
2. Vois AI chunks the content, generates vector embeddings, and stores them in a search index
3. When an agent receives a question, it runs a semantic search against the index
4. The top matching passages are injected into the agent's context before it responds

Semantic search means the agent finds relevant content even when the user's wording doesn't match exactly — "how do I connect Google Calendar" will match a page titled "Calendar Integration Setup."

Adding content

Open the agent editor → Knowledge Base tab. There are three ways to add content:

Upload files

Drag and drop files directly onto the upload zone, or click to browse. Supported formats:

Format	Notes
PDF	Text is extracted from all pages
TXT	Plain text, ingested as-is
MD	Markdown files (great for docs exports)

Paste text

Click Paste text to open an inline editor. Name the document and paste any text content — useful for FAQs, policy text, or anything you don't have as a file.

Connect a URL (docs sites, wikis, help centres)

Click Connect URL and enter a website URL. Vois AI will:

1. Fetch sitemap.xml to discover all pages automatically (falls back to link crawling if no sitemap)
2. Strip navigation, footers, and scripts — extracting only readable content
3. Chunk and embed each page into the knowledge base
4. Show status as Processing → Active as pages are indexed

tip

This is the recommended approach for documentation sites. Just enter the root URL of your docs (e.g. https://docs.mycompany.com) and all pages are indexed automatically — up to 80 pages per source.

URL sources are shown with a globe icon in the document list. To re-index after updating your docs, delete the source and re-add the URL.

Document status

Status	Meaning
Processing	Content is being chunked and embedded — may take a few seconds to a minute
Active	Indexed and searchable by agents
Failed	Ingestion failed — check that the URL is publicly accessible or the file is readable

Enabling on an agent

The knowledge base is automatically available to any agent assigned to the project when RAG is enabled. The agent uses it through a built-in search tool — no extra configuration needed.

To use knowledge base search in a skill:

Search your knowledge base for relevant documentation before answering.
If you find a relevant answer, share it and ask if it resolves the question.
Only escalate to a support ticket if the knowledge base doesn't contain the answer.

See Support Ticket Routing for a full docs-first support skill.

Best practices

• Use URL sources for live docs — they reflect your current content without manual re-uploads
• Structured content retrieves better — clear headings, short paragraphs, and lists chunk more cleanly than dense prose
• Upload focused documents — a specific FAQ file retrieves more precisely than a large omnibus document
• Test in the playground — after adding content, ask questions that should be answered by the docs and verify the agent finds them
• Keep sources current — outdated content can mislead the agent; delete and re-add URL sources when docs change significantly