Q&A pairs#

Explicit question/answer pairs that take precedence over RAG retrieval. The closest thing Chatvia has to a "scripted reply". Useful when your team needs the agent to respond with exact, approved wording.

When to use a Q&A pair#

• A canonical question the agent gets a lot, where your team wants a deterministic answer.
• A legal-sensitive answer where wording matters (privacy, billing, regulated industries).
• A fallback for an FAQ topic where the model's free-form answer keeps drifting.

A few pairs go a long way. Your team doesn't need to script everything.

Adding a Q&A pair#

In Sources → Q&A, click Add Q&A:

• Title: an internal label, like "Refund requests".
• Questions: the ways a customer or teammate would ask it. Click Add question for each extra phrasing.
• Answer: the exact response the agent will use (at least 10 characters).

The questions are what gets embedded and matched. The answer is served back word for word. So:

• Write each question the way a chatter would say it.
• Write the answer the way your workspace wants the agent to reply.

How retrieval prefers Q&A#

Before it searches the other sources, Chatvia checks the chatter's message against your saved questions. When the message is the same as a saved question (ignoring case, accents and punctuation), or very close to one in meaning, Chatvia sends the saved answer exactly as written. No AI model writes the reply, so the answer is instant and always the same.

If nothing matches closely enough, retrieval falls back to other sources as usual. Q&A pairs are special has the details.

Several phrasings#

Adding several phrasings of the same question to one pair improves the match rate. For example, one pair can hold:

Q: How do I cancel my subscription?
Q: Where do I unsubscribe?
Q: I want to stop my account
A: You can cancel from Settings → Billing. Your access continues until the end of the current billing period.

Editing#

After you save, the pair is ingested again, usually in seconds. Then the new wording is used across every surface (Hosted Chat, Widget, Bar).

Q&A vs. Text vs. Files#

Rule of thumb:

• Files / Web: broad knowledge the model should reason over.
• Text: short content the model should reference.
• Q&A: specific questions your team wants answered the same way every time.