Knowledge base

A knowledge base is the content your agent draws on during a call so it answers from your own material instead of guessing. Voxy supports three source types — URLs, files, and raw text — all turned into the same searchable format.

Source types

• URL — we fetch the page, strip the navigation and boilerplate, and keep the readable content.
• File — PDF, DOCX, TXT, or Markdown. We extract the text automatically.
• Text — paste content directly in the editor at /workspace/knowledge-base.

How content is prepared

1. Extract — we pull clean text from each source.
2. Split — long content is split into overlapping passages so context carries across boundaries.
3. Index — each passage is indexed for fast, meaning-aware lookup during a call.

Retrieval at call time

While the agent is on a call, it looks up the most relevant passages for what the caller is asking and grounds its answer in them. Voxy blends two kinds of matching:

• Keyword — matches the exact words and names in your content.
• Meaning — matches by intent, so the agent still finds the right passage when the caller uses different words.

Combining the two keeps the agent accurate on synonyms (“billing” vs “invoice”) as well as on exact product names and terms unique to your business.

Re-ingest

URLs and files can be refreshed in place — the old content is replaced cleanly. Useful whenever your help-centre page changes.