Limits and quotas

The server sends the binding values for a connection in init_ok when the connection is established; the widget sticks to them. The plans themselves are listed on the pricing page.

Messages

• Length of a visitor message: 1000 characters on the Free plan, 4000 characters on paid plans (counted in Unicode code points). The widget does not send longer texts; otherwise the server responds with msg_too_long. Messages from your team may always be 4000 characters long.
• Channels (email, WhatsApp …): longer inbound messages are shortened to the limit and marked with [...].
• Server API and MCP: at most 4000 characters per message.
• Rating comment: 1000 characters. Prefilled text via data-rc-message/open: 200 characters.

File attachments

• Free: at most 2 MB per file, retention 7 days, 1 GB per workspace.
• Paid plans: at most 15 MB per file, retention 90 days, 50 GB per workspace.
• Negotiated contracts: as per contract, at most 32 MB per file.
• Per message: at most 5 attachments.
• Allowed types: PNG, JPEG, WebP, HEIC/HEIF, SVG, PDF, text (text/plain) and CSV. The server checks the file header; if it does not match the declared type, the file is rejected (upload_mime_denied or HTTP 415).
• Upload approval: valid for 5 minutes (after that upload_expired; the widget requests a new one).

The applicable values are in init_ok.features.max_upload_bytes and allowed_mime.

AI assistant

• Per conversation: at most 8 AI requests per minute (Free), 15 (Professional), 30 (Agency).
• Per project: at most 30 AI requests per minute across all visitors (Free), 60 (Professional), 120 (Agency and contracts).

Above that, the assistant temporarily accepts no further AI requests in this conversation or project until the minute has passed.

Chat connection

• Frame: at most 64 KiB.
• Per connection: 20 messages and 120 frames per 10 seconds (token bucket), above that rate_limited.
• Per session: at most 3 simultaneous connections (several tabs share one anyway).
• Keep-alive: ping every 25 seconds; after 70 seconds without a frame, the server closes the connection.
• Reconnecting: delay up to 30 seconds.
• Deduplication: 24 hours.
• Offline buffer in the browser: 50 messages, 128 KiB, 24 hours.
• Abuse protection per address: connection attempts, simultaneous connections and frames per second are limited; repeated violations lead to a block (close 1008). Behind your own proxy, see Own domain.

More under Transport.

HTTP fallback

• Long polling holds for up to 25 seconds, keep-alive in the stream every 15 seconds.
• Sending: at most 20 frames per request.
• After 60 seconds without a request, the connection is closed.

Server API

• 120 requests per minute and key (X-RateLimit-*, Retry-After).
• Body at most 64 KiB, limit at most 100.
• Events (POST /v1/events): payload at most 8 KiB.
• REST hooks: at most 200 per workspace.
• MCP kb_suchen: 20 searches per minute and key.

More under Server API.

Webhooks

• Timeout of 4 seconds for the connection setup, 10 seconds in total.
• Up to six delivery attempts (1 s, 5 s, 25 s, 125 s, 625 s); events older than one hour are discarded.

Customer data

• Fetch: response within 3 seconds, at most 16 KiB, 10 orders, 20 items, texts 200 characters.
• Import: at most 50 orders per customer, deleted 90 days after the last import.

Smart Answer

• 60 requests per minute and visitor address.
• Question 1–300 characters; body 4 KiB (HTML box) or 8 KiB (answer stream).
• Without text after 45 seconds or without an end after 90 seconds, the answer is aborted.

Plan limits

Conversations per month, projects and team members depend on your plan (Pricing).