sitegpt knowledge.
For step-by-step tasks, use the task pages:
- Agent-first onboarding
- Install and log in
- Authentication
- Chatbots
- Knowledge
- External data sources
- Customization
- Conversations
How to read this reference
- Tables list commands in the form you type them. Placeholders are in angle brackets, for example
<chatbot-id>. - Aliases are other names for the same command. They are listed in the description.
- A command shown with
--yesfails without it. See Output and exit codes. - Every command level has help, for example
sitegpt --help,sitegpt knowledge --help, orsitegpt settings appearance --help. - Use
--jsonwhen you need full IDs, cursors, nested data, or output for scripts.
Global options
You can use global options with any command.
Environment variables:
Other rules:
Output and exit codes
Agent-first onboarding
For users who have no SiteGPT account yet, or who want to preview a chatbot before signup.onboarding start needs no login. The other commands need the temporary token that onboarding start returns. For the full flow, see Agent-first onboarding.
Login
Login saves a token in a local profile and makes that profile the default. With no--profile, the profile is default. sitegpt login uses the OAuth device flow. For the flow, scopes, and access levels, see Authentication.
--token. See Authentication.
Logout
Logout removes a local profile from your machine. It does not revoke the token in SiteGPT. To revoke it, usesitegpt tokens revoke <token-id>.
Whoami
Profiles
Profiles are named credentials saved on your machine.API tokens
API tokens are scoped credentials for the CLI and AI agents. For the list of scopes, see Authentication.Account
Usage
Limits
Billing
Billing commands only read data. They do not change your plan.Chatbots
Most chatbot commands need--chatbot <chatbot-id>.
Dashboard summary
The summary includes the training state, document counts, content counts, message count, positive and negative feedback percentages, the chat URL, and the widget script URL.
Engagement analytics
Installation
Icons
Knowledge documents
A document is one item of content that SiteGPT trained on, such as one page or one file.knowledge docs is an alias for knowledge documents.
Source values:
--state group that each belongs to:
Type values:
Shared options for adding web pages
knowledge links add, knowledge website add, and knowledge sitemap add all accept these options.
Knowledge links
Also accepts the shared options for adding web pages.
Knowledge website
Also accepts the shared options for adding web pages.
Knowledge sitemap
Also accepts the shared options for adding web pages. Frequencies that the plan does not include are set to
NEVER with a warning. If sync and scan use the same frequency, only the sync runs on that schedule. See Keep your content up to date.
Knowledge YouTube
Knowledge text
Each chatbot has one text snippet. These commands replace the whole snippet.
The text can be up to 10,000 characters.
Knowledge files
Knowledge sync jobs
A sync job holds the sync and scan schedule of a source. Auto-sync runs only for website, sitemap, and GitHub sources. Scan runs only for sitemap sources. The list hides jobs with both frequencies set toNEVER. knowledge jobs is an alias for knowledge sync-jobs.
update needs --sync, --scan, or both.
External data sources
Connect apps such as Google Drive or Notion, and ingest the files you select.knowledge connections is an alias for knowledge sources. For the steps, see External data sources.
update needs at least one option.
GitHub source helpers
Confluence source helpers
Custom responses
A custom response is a question and the answer you want the chatbot to use for it. See Add custom responses.update needs --question, --answer, or both.
Personas
A persona sets the chatbot’s tone and style. The chatbot uses the persona you select withuse. See Write instructions.
Instructions
Instructions tell the chatbot how to answer. Each set of instructions has its own temperature. The chatbot uses the set you select withuse. See Write instructions.
Settings overview
You can read all settings at once, update them from a JSON file, or change one section at a time. For what each dashboard setting does, see Chatbot settings.Settings: general
When the plan of the chatbot owner is below Growth, the
--rate-limit-thread-* options fail with RATE_LIMITS_NOT_AVAILABLE. Your own plan does not count. See Plans and limits.
Settings: appearance
Text and color options:
Layout options:
On and off options. Each takes a
<boolean> value, for example --dark-mode true. All default to false.
Plan-dependent options fail if the plan does not include the feature. The watermark options fail with
WATERMARK_SETTINGS_NOT_AVAILABLE. The CTA options fail with CTA_SETTINGS_NOT_AVAILABLE. See Plans and limits.
Settings: chat mode
set turns working hours and visitor choice off and keeps their settings. To set working hours or visitor choice, use the dashboard or PATCH /chatbots/{chatbotId}/settings/chat-mode in API v2. See Answer during working hours and Let the visitor choose the AI or a person.
Settings: localization
Settings: user data
Pre-chat user details. Off by default. In the dashboard, this is Pre-Chat User Details in Leads > Settings. See Collect leads.
Email is always collected.
--collect-name and --collect-phone change the same stored values as the lead form options with the same names.
Settings: lead form
To turn on Email for every form submission, use
--file with {"notifications": {"everySubmission": true}}. Do not add --notifications-enabled or --notification-email to the same command. These flags replace the notifications object from the file.
Settings: human support
The dashboard field When should the AI escalate? has no flag. Set the
escalationPolicy field with --file. It takes up to 2,000 characters. Leave it empty to use the default triggers.
Settings: webhooks
If the chatbot owner’s plan does not include webhooks, these commands fail with
WEBHOOKS_NOT_AVAILABLE. See Webhooks and Plans and limits.
Conversation starters
Conversation starters are buttons that visitors see before they send a message.
Conversation starters cannot be
ESCALATION buttons.
Conversation followups
Follow-up suggestions are buttons shown after the chatbot answers. The CLI calls themfollowups.
--page and --clear-pages work only for conversation starters.
Conversations
A conversation is the chat between one visitor and your chatbot or team. Commands use the conversation ID as<thread-id>. For the steps, see Conversations.
For the error codes, see API conventions. For how modes and escalation work, see Conversations and human handoff.
Messages
Messages are the entries in a conversation. The CLI sends every message as the visitor.Tags
Tags group conversations. Each tag belongs to one chatbot, so every tag command needs--chatbot <chatbot-id>.
To assign tags to a conversation, use
conversations update --tag.
Leads
A lead is the contact details that a visitor gives in the chat.Members
Members are team members who can access a chatbot. See Team members.Member invites
MCP server
It uses the same token as the CLI:
SITEGPT_API_TOKEN or a saved profile. For setup, the hosted server at https://sitegpt.ai/mcp, and the npx -y @sitegpt/mcp launcher, see MCP.
In a HIPAA workspace, sitegpt mcp does not return conversations or leads. See MCP server.
Agent guide
The guide links to the full skill file at
https://sitegpt.ai/agents/sitegpt-cli-skill.md.