Before you put your agent in front of real customers, test it. Vosaire has a built-in chat window that lets you talk to your agent exactly as a user would.
Opening the test window
From any agent detail page, click the Test Agent button in the top right corner. A chat panel opens on the right side of the screen.
This is a live session — the agent is using the same knowledge sources and behavior settings it would use for a real customer. Any changes you make to settings will affect the next message you send in the test window.
What to test
Test the easy questions first. Pick 3–5 things your content clearly covers and ask them as a real user would:
- "How do I reset my password?"
- "What payment methods do you accept?"
- "Do you offer a free trial?"
If the agent handles these well, move on to harder ones.
Test the edge cases. Ask things that are only partially covered, or that require combining information from multiple sources:
- "What happens if I downgrade my plan mid-month?"
- "Can I use this on mobile?"
Test what it doesn't know. Deliberately ask something not in your knowledge base. A well-configured agent should say it doesn't know and offer to help in another way, not make something up.
What good looks like
- Accurate answers drawn from your actual content (you'll sometimes see source citations)
- Natural language, not robotic
- Appropriate handling of unknowns ("I don't have that information, but you can reach our team at…")
- Correct behavior per your instructions (right tone, correct escalation behavior, etc.)
What to do if something's wrong
Agent making things up? It's either guessing because sources don't cover the topic, or the instructions are too vague. Add a specific FAQ entry for that question, or add a line to your instructions: "Only answer from provided knowledge. If unsure, say you don't know."
Agent giving generic answers? Usually means the relevant content wasn't indexed properly. Check the source status (it should say "Indexed", not "Error"). Try re-adding the source.
Wrong tone or style? Go to the Behavior tab and edit the instructions. Be specific — "Keep responses under 100 words" works better than "be concise."
Ignoring part of the question? Try rephrasing your sources so the relevant answer is clearer and more direct.
Iterating quickly
The test window is the fastest feedback loop you have. Make a change to behavior or add a source, then immediately re-test the question that was failing. You can have both the Behavior/Knowledge tab and the test window open at the same time — just click between tabs.
Don't over-optimize before going live. Get it to a point where it handles the top 10 most common questions well, then launch. Real conversations will show you what to improve next — use the Train tab to correct responses based on actual user interactions.