Frequently asked questions
Short answers to what people ask before and just after they start.
About the agent
Does the agent make things up?
It is built not to. It may only answer from your sources and from facts in your instructions, and it is told to say so when they do not contain the answer. No language model is perfect at this, which is why every answer is measured twice. Answers that went beyond their sources are flagged for you on the Overview. Confidence and grounding.
Will it answer questions that have nothing to do with my business?
No. It is not a general-purpose assistant, and it declines to write poems, code or homework. If you want it to do more, say so in the instructions.
Does it learn from conversations?
No. It knows what is in its sources, and nothing else changes it. You improve it by adding sources, which keeps you in control of what it says.
Is my content used to train AI models?
Not by us. We train no models. Data and privacy lists which providers process what.
How much of a conversation does it remember?
The last 10 messages. A visitor can scroll back further than that on their screen, but the agent no longer takes older messages into account.
Which languages does it speak?
The widget's own texts exist in English, Dutch, French, German and Spanish. The agent answers in the language the visitor writes in and switches with them without asking, also when your sources are in a different language. Languages.
Can it hand a conversation over to a person?
Not live. It can refer a visitor to a phone number or an e-mail address from your instructions, and it can let them book an appointment inside the chat. With tickets switched on, a visitor can also leave their question for your team in the chat, and your team answers by e-mail.
About sources
How often is my website re-read?
It is not, until you do it. What the agent knows is what it read on the day you added the site. Limits and freshness.
Can it read a page behind a login?
No. The crawler reads what any anonymous visitor can see. Put such content in a file or a text source.
Can it read scanned documents or images?
Not as sources: they are not read by text recognition (OCR), and a scanned PDF fails, with the message that no text was found in it. Files. From a visitor, yes: a scan they send is read by text recognition, and a picture is looked at on a model that reads images. Files from visitors.
Can I connect Google Drive, Notion or a database?
Not directly. Export the content as files or text, or add it with the API.
About the widget
Does the widget slow my site down?
Hardly. The script only draws the button. The chat itself loads when a visitor opens it. Install.
Does it set cookies? Do I need to change my cookie banner?
It sets no cookies, and stores nothing until the visitor sends a message or a file. Then it keeps two small values in the browser's storage to hold that conversation together until the tab is closed. Neither identifies a person. The one exception is yours to make: a site that passes its visitors' consent lets the button keep one value for the visit beforehand, to time the second nudge. Whether that needs mention in your cookie notice depends on your own policy. Most treat it as functional.
Can I remove "Powered by ChatterLab"?
Yes, with the badge removal add-on. The line saying that the visitor is talking to an AI can never be removed. Appearance.
Can visitors send files or images?
Yes, from Studio up: PDF with text, Word, Excel, PowerPoint, OpenDocument and plain text, and a photo or a screenshot when the agent's model reads images. Files from visitors.
About plans and paying
What happens when my credits run out?
The agents stop answering until the next month, unless you buy credits or have switched on auto-recharge. Plans and credits.
Do test questions cost credits?
Yes. Every answer costs credits, wherever the question came from.
Can I work in one workspace with colleagues?
Not yet. A workspace belongs to one account.
Can I move an agent to another account?
Not yourself. Write to management@sri-group.eu.
For developers
Is there an API?
Yes, for everything the dashboard does, on Studio and up. API overview.
Can I build my own chat interface?
Yes, with POST /agents/{id}/chat, streaming or not, from your own server. Remember that telling users they are talking to an AI is then up to you. The EU AI Act.
Can Claude or another assistant manage my agents?
Yes, over MCP, on Studio and up.