How is the webhook payload signed?+
Every webhook ships with an X-Agentbot-Signature header — HMAC SHA-256 over the raw request body, signed with your secret. Three lines on your end to verify (HMAC the body, constant-time compare to the header). The payload shape and signature scheme are documented at /docs/webhooks.
Can the AI read our docs site directly?+
Yes — enable web search and provide your docs URL. The AI fetches relevant pages live on every conversation turn. Combine with uploaded knowledge base entries for internal-only context (architecture, support runbooks, deprecation notes) that isn't on the public docs.
Does the AI know how to write code in replies?+
Yes, with sensible formatting. The system prompt is configurable — set the tone to 'developer audience' and the AI emits code fences with language tags, links to specific doc paths instead of just doc homepages, and drops the marketing-speak. The bundled markdown rendering displays code blocks correctly in the widget.
Can it file issues directly into Linear or GitHub?+
Via webhook. When the AI files a ticket, the ticket-created event fires; pipe it to Linear's API (via Zapier, n8n, or direct) to create an issue. Same for GitHub. We don't ship native Linear / GitHub integrations yet — webhooks cover the use case in 20 minutes of setup.
What about self-hosting?+
Agentbot itself is hosted (Vercel + Supabase). We don't currently offer a self-hosted distribution. The widget is one script tag and the data export gets your data out at any time, so the lock-in surface is small — but if you must self-host, we are not the right pick today.
Is there an API for managing sites / conversations programmatically?+
Read-side API for conversations and tickets is on the roadmap. Today, the webhook surface covers most read-side needs (subscribe to events and stream them into your warehouse). For write operations (create a ticket via API, post a message via API), the API is being designed; reach out via /contact if you have a specific use case so we can prioritise.