Before you start
- An AI agent that can run terminal commands, or a terminal of your own.
- Node.js 18 or later, with npm.
- The URL of the website the chatbot should learn from.
- No SiteGPT account or plan is needed to start. To claim the chatbot, you choose a plan at checkout.
Build and share the chatbot
1
Install the CLI
2
Start onboarding from the website URL
3
Save the values from the response
4
Use the temporary token
Set the token in the shell where the agent works:The CLI now uses this token for every command in this shell.
5
Set up the chatbot
Use the normal CLI commands to add content, a persona, instructions, appearance, conversation starters, follow-up suggestions, and icons. For example:For a full setup sequence, see New customer onboarding.
6
Check setup progress
data.setupChecklist. Fix items that are PENDING or WARNING before you share the chatbot. The states are listed in Setup checklist states.7
Share the onboarding URL
Give the person the
data.onboardingUrl value. On that page they can test the chatbot and claim it.Check that it works
sitegpt onboarding status YOUR_WORKSPACE_IDshows a Setup checklist with the items marked Done.- The onboarding URL opens a page where you can chat with the chatbot.
Claim the chatbot
After the person tests the chatbot, ask if they want to keep it. If yes, ask for these values:
Then run:
Delete an unclaimed chatbot
If the setup is wrong or the person does not want the chatbot, run:--yes.
Reference
Setup checklist states
The checklist covers content, persona, instructions, conversation starters, follow-up suggestions, and brand styling.Temporary token
Agent discovery
Agents that readhttps://sitegpt.ai/auth.md find the same no-account flow there. See Agent discovery reference.