1. Choose the conversation you want to support.
Tell us whether your first use case is product support, sales questions or both. Pilot access is arranged with the team. Start with a specific audience and a short set of questions so you can judge whether the answers are useful.
2. Give the agent sources your team has checked.
Add support FAQs or selected GitHub documentation. For sales, add manual product, feature and pricing notes. Review and approve each source in the dashboard before it becomes available to the appropriate agent.
3. Set the greeting and test the answers.
Choose the agent’s name, voice, greeting and instructions. Try representative questions, inspect the transcripts and revise the knowledge where needed. Tests use the same workspace conversation allowance as customer calls.
4. Install the widget and review conversations.
Add your website to the allowed origins and copy the generated embed into the site. Enable customer calls after testing. The widget collects the caller’s topic, requests microphone permission and opens the voice conversation.
Use the dashboard to review transcripts and feedback. Keep the knowledge current and reapprove changed GitHub sources before the agent uses them.
Common questions
Can I set it up through an AI coding agent?
Yes. The product exposes a workspace-scoped management API and an authenticated MCP endpoint for trusted agents. They can add knowledge, approve sources, update agent settings, inspect conversations and get installation details.
Will submitting the pilot form install anything on my site?
No. The form helps you enquire about a pilot, either by preparing an email for you to send or through connected submission when available. Installation and access are discussed with you before any widget is added to your website.