Build a preview chatbot before the user has an account

Use this playbook when the user has no SiteGPT account, gave you no token or profile, or asks to try SiteGPT for a website. You build a temporary chatbot with no login. Then you share one onboarding URL. On that page the user previews the chatbot and claims it.

Before you start

Account: none needed. sitegpt onboarding start does not need a login.
CLI: installed. See Install and log in.
Tools: curl, grep, and file, to read the website and download the logo.
Input: the website URL and the chatbot purpose. See Ask for the purpose first.

Do not run sitegpt whoami or sitegpt login for this flow. If a command returns PROFILE_NOT_CONFIGURED, that is expected. Continue with the temporary token.

Steps

Start the onboarding workspace

sitegpt onboarding start https://example.com --agent-name "<agent-name>" --json

Save these fields from the response:

Field	Use
`data.workspace.id`	Workspace ID for `onboarding status`, `claim`, and `delete`.
`data.workspace.chatbotId`	Chatbot ID for every setup command.
`data.apiToken`	Temporary token. It works only for this chatbot.
`data.onboardingUrl`	The page where the user previews and claims the chatbot.

WORKSPACE_ID=<data.workspace.id>
CHATBOT_ID=<data.workspace.chatbotId>
export SITEGPT_API_TOKEN=<data.apiToken>

With SITEGPT_API_TOKEN set, every later command uses the temporary token. Do not show the full token to the user.

Read the website

Use a browser or fetch tool to read the text. Use raw HTML to find colors, icons, and the sitemap.

curl -sL https://example.com -o /tmp/sitegpt-page.html
grep -oE '#[0-9a-fA-F]{6}' /tmp/sitegpt-page.html | sort | uniq -c | sort -rn | head -30
grep -oE 'name="theme-color"[^>]*' /tmp/sitegpt-page.html
grep -oE '<link[^>]*rel="[^"]*icon[^"]*"[^>]*' /tmp/sitegpt-page.html
grep -oE '<link[^>]*apple-touch[^>]*' /tmp/sitegpt-page.html
grep -oE 'property="og:image"[^>]*content="[^"]*"' /tmp/sitegpt-page.html
curl -sL https://example.com/robots.txt
curl -I https://example.com/sitemap.xml

Write down the brand name, what the product does, the main support topics, the contact email, the sitemap URL, the brand color, and the best square logo. Do not guess a color or invent facts.

Add content

Use the sitemap if the site has one. It usually gives the cleanest coverage.

Sitemap
Website crawl
Selected links

sitegpt knowledge sitemap add \
  --chatbot $CHATBOT_ID \
  https://example.com/sitemap.xml \
  --only-main-content true \
  --json

sitegpt knowledge website add \
  --chatbot $CHATBOT_ID \
  https://example.com \
  --depth 3 \
  --max-links 100 \
  --only-main-content true \
  --json

sitegpt knowledge links add \
  --chatbot $CHATBOT_ID \
  https://example.com/pricing \
  https://example.com/docs \
  https://example.com/contact \
  --only-main-content true \
  --json

Upload the logo

Download the logo to a local file. Then upload it as the bot icon and the chat bubble icon.

curl -sL https://example.com/logo-icon.png -o /tmp/sitegpt-logo.png
file /tmp/sitegpt-logo.png
sitegpt icons upload --chatbot $CHATBOT_ID bot /tmp/sitegpt-logo.png
sitegpt icons upload --chatbot $CHATBOT_ID chat-bubble /tmp/sitegpt-logo.png

If the favicon is small or blurry, use a clean logo file from the site.

Set the appearance

Use the brand color from step 2. Pick a text color that is easy to read on it.

sitegpt settings appearance update \
  --chatbot $CHATBOT_ID \
  --title "<Brand> Support" \
  --welcome "Hi! I can help with <Brand> questions." \
  --placeholder "Ask about <Brand>..." \
  --brand-color "#0F766E" \
  --brand-text-color "#FFFFFF" \
  --link-color "#0F766E" \
  --icon-shape CIRCLE \
  --icon-position RIGHT

Create and select a persona

Write the persona to a file. Match it to the purpose. For support, stress accuracy and clear next steps. For lead generation, stress short product answers and a clear call to action. For docs, stress answers that come from the content.

cat > /tmp/sitegpt-persona.md <<'EOF'
You are the <Brand> support assistant.
Use a clear, friendly, and practical tone.
Answer from the chatbot content first.
Offer the most useful next step.
EOF

sitegpt personas add \
  --chatbot $CHATBOT_ID \
  --title "<Brand> support specialist" \
  --file /tmp/sitegpt-persona.md \
  --json
sitegpt personas use --chatbot $CHATBOT_ID <persona-id>

Create and select instructions

cat > /tmp/sitegpt-instructions.md <<'EOF'
Answer from the chatbot content first.
If the content does not have the answer, say so and suggest how to contact support.
Do not invent prices, policies, legal terms, availability, or integrations.
Keep answers short unless the visitor asks for detail.
Ask a clarifying question when the request is not clear.
EOF

sitegpt instructions add \
  --chatbot $CHATBOT_ID \
  --title "Grounded <Brand> assistant" \
  --file /tmp/sitegpt-instructions.md \
  --temperature 0.3 \
  --json
sitegpt instructions use --chatbot $CHATBOT_ID <instruction-id>

Add conversation starters and follow-up suggestions

Use questions a visitor would click. They also make the preview easy to test.

sitegpt starters add --chatbot $CHATBOT_ID --title "What does <Brand> do?" --message "What does <Brand> do?" --json
sitegpt starters add --chatbot $CHATBOT_ID --title "Pricing" --message "Tell me about pricing and plans." --json
sitegpt starters add --chatbot $CHATBOT_ID --title "Getting started" --message "How do I get started?" --json
sitegpt followups add --chatbot $CHATBOT_ID --title "Contact support" --message "How can I contact support?" --json
sitegpt followups add --chatbot $CHATBOT_ID --title "Compare options" --message "Can you help me choose the right option?" --json

Set lead or support options, if the purpose needs them

Do this only when the website or the user gives you the email address.For lead generation:

sitegpt settings lead-form update \
  --chatbot $CHATBOT_ID \
  --enabled true \
  --collect-name true \
  --notification-email sales@example.com

For customer support:

sitegpt settings general update \
  --chatbot $CHATBOT_ID \
  --support-email support@example.com

Fix the setup checklist

sitegpt onboarding status $WORKSPACE_ID --json

Each checklist item is DONE, PENDING, WARNING, or UNKNOWN. Fix the items that are not DONE when you can.

Check that it works

Wait until the content finishes training:
sitegpt knowledge wait --chatbot $CHATBOT_ID
The Training settled line reports how many documents trained and how many failed.

Send realistic visitor questions:

sitegpt messages send --chatbot $CHATBOT_ID "What does this company do?" --json
sitegpt messages send --chatbot $CHATBOT_ID "How much does it cost?" --json
sitegpt messages send --chatbot $CHATBOT_ID "How can I contact support?" --json

Each answer should use facts from the website.

Run sitegpt onboarding status $WORKSPACE_ID again. Each item that you could fix now shows DONE.

Hand back to the human

Share the onboarding URL (data.onboardingUrl) only after the chatbot gives useful answers. Tell the user:

What the chatbot is for, and any purpose you inferred.
Which content you added, and how many documents failed.
The brand color and icon you used.
The onboarding URL, where they can preview and claim the chatbot.

If the user wants to claim the chatbot from the CLI, ask for their email, plan, and billing interval:

sitegpt onboarding claim $WORKSPACE_ID \
  --email user@example.com \
  --plan GROWTH \
  --interval MONTH \
  --json

Plans are STARTER, GROWTH, and SCALE. Intervals are MONTH and YEAR. If the response has data.checkoutUrl, give that link to the user. If the email already has a SiteGPT subscription, ask the user to open the onboarding URL while signed in and claim the chatbot there. If the user does not want the chatbot, delete the workspace:

sitegpt onboarding delete $WORKSPACE_ID --yes

Get started

Chatbot setup

Operations

Agent playbooks

Reference

Build a preview chatbot before the user has an account

Before you start

Steps

Check that it works

Hand back to the human

​Before you start

​Steps

​Check that it works

​Hand back to the human

Before you start

Steps

Check that it works

Hand back to the human