Start in your website project
This is the entire prompt. Your agent discovers the setup instructions from stand.chat.
Add stand.chatThe agent installs the official @standchat/cli package from npm’s latest tag. If the package is unavailable, it stops and reports the missing distribution. It must not substitute another package or ask for pasted access tokens.
01
Ask your coding agent to add Stand
Open the website project in a coding agent that can edit files, run shell commands, and use a real browser. Say “Add stand.chat”. The agent finds its setup instructions on the Stand homepage or in llms.txt; you do not need to supply a guide URL or know the installation steps.
The CLI runs on macOS, Linux, and WSL with Node.js 22 or later. Native Windows is not supported. The agent needs network access and permission to edit and publish the intended website.
02
Complete signup and consent yourself
The agent starts CLI login and gives you a browser verification link and code. Create an account or sign in, review the requested permissions, and accept the terms and consent yourself. You do not give the agent your password or browser cookies.
If your Stand profile is incomplete, the agent opens the existing browser onboarding form. Complete it with the same account used for CLI login. The CLI reads whether onboarding is complete; it does not create your profile or organization.
03
Let the agent configure your chat
The agent checks your plan and existing configuration, creates a missing Site and a disabled Stand-in, and adds approved knowledge. Base supports this setup through Stand-in instructions and website crawling within its normal limits.
Choose what visitors should be able to learn. Public website content can be crawled. If your account allows document uploads and ingestion, the agent can upload approved local documents without publishing them on your website. Private project material needs your specific approval and must suit the chat audience. An unpublished website can also start with approved facts in the Stand-in instructions. Unfinished ingestion is reported as unfinished.
04
Enable and install the widget
The agent enables the configured Stand-in and installs the Site’s returned script snippet once, using the website framework’s supported script lifecycle. Credentials stay in the CLI store and out of your website code.
A shared Stand-in coverage switch can affect other enabled Stand-ins. The agent checks that before changing it and asks you to approve effects beyond this new chat. It leaves human availability unchanged and uses only authorized deployment access.
05
Check an actual visitor answer
The agent opens the intended website in a real browser, starts a fresh visitor chat, and asks about an approved fact. It checks that the intended Stand-in answers and acknowledges when information is missing.
A greeting, script tag, installed flag, or queued crawl is not enough. The final report identifies the website, configuration and code changes, knowledge result, and chat verification. Missing deployment access or an unverified answer means setup is still incomplete.
Keep control of an existing setup
The agent reads existing resources before creating anything and keeps the IDs returned by successful operations. After an interruption, it checks those resources before continuing. If a write’s outcome is uncertain, it stops for reconciliation instead of blindly repeating it.
The API and CLI also support ongoing management with separately granted permissions. Account deletion, billing and team membership changes stay in the browser. Website JavaScript experimentation requires the owner’s browser opt-in. MCP and WordPress plugins are outside this workflow.