Chat guardrails

TL;DR

Rolling out. Chat guardrails limit how many messages one visitor can send, pause obvious spam and block countries you choose, so abuse of your widget does not burn your message credits. Settings save now and apply once enabled.

Guardrails are the abuse controls for visitors chatting with your agent on your website. They live in Settings → Security → Chat guardrails, and they apply per workspace.

Note

Guardrails are rolling out and are off by default. You can save your settings now; the settings page tells you plainly when they are not yet active on your widget. Nothing here is enforced until the flag is on for your workspace.

Rate limit

Limit how many messages one visitor can send in a given window: for example, 5 messages every 30 seconds. When a visitor goes over, they get the message you wrote, as a normal reply in the chat, and that turn does not use any of your message credits.

• Messages: 1 to 1000.
• Every (seconds): 1 to 86400 (24 hours).
• Message shown when the limit is hit: up to 500 characters. Write it in your own voice; the visitor sees exactly this text.

The window is counted per visitor device, not per site, so one abusive visitor cannot lock out everyone else on your storefront.

Spam detection

Turn this on and describe, in your own words, what counts as spam for your agent: for example "crypto promotions, link dumps, repeated nonsense". You get up to 2,000 characters.

The check runs after a visitor's 2nd, 4th, 8th and 16th message, not on every turn, so it costs almost nothing. When the conversation is judged to be spam, the AI is paused for that conversation and the visitor is told the conversation has been paused. The conversation stays in your inbox, and you or a teammate can still reply by hand.

If our classifier is unavailable for any reason, the conversation is allowed through. We would rather serve a spammer than silence a real customer.

Blocked countries

Enter the two-letter ISO 3166-1 alpha-2 codes for countries you do not want to serve: for example RU, CN, KP. Visitors from those countries get a short "not available in your region" reply instead of an answer.

What this covers: the chat bubble on your website and your hosted chat page. What it does not cover: WhatsApp, Telegram, Slack, SMS and other messaging channels: those carry no reliable country signal, so we do not pretend to filter them.

Visitors whose country we cannot determine are served, not blocked. We never guess a country from an IP address.

What guardrails are not

Guardrails are abuse controls, not a security boundary. They reduce nuisance traffic and runaway credit spend; they are not a substitute for restricting which domains your widget may be embedded on (Settings → Web widget → Allowed domains).

Related guides

• Widget setup, for the chat these limits protect.
• Usage meter, to watch message credits.
• AI safety, for how answers stay grounded.

Ready to ship this on your site?

Free plan with 50 messages a month and 5 MB of content. No credit card, no time limit.

Try AskVault Free → See pricing

Was this page helpful? Yes No