Quickstart

Check your Node

node --version

Anything below 22.19.0 and the launcher stops before it loads anything, printing the version you have, the installer link and the install command for your platform.

Run the wizard

npx @chatfuel/wizard

It has questions, so it needs a terminal. Piped, redirected or started by a CI job, it stops before the first one and says so; --yes with the answers as flags is the unattended form.

Answer its questions

How should the modules be installed? New app. The other answer copies the modules into a project you already have — that is the embed guide.

Which modules do you want? Fifteen exist. core comes with all of them and is not in the picker; the fourteen it lists are livechat, contacts, deals, bookings, knowledge-base, automations, coworker, flow-builder, ads-optimization, publishing, broadcasts, channels, auth and admin — the same ids --modules takes. auth and admin are marked opt-in in the picker: both need credentials the wizard cannot invent, so a --yes run leaves them out unless you name them. Pick deals and contacts comes with it, named as it is added; if a pick recommends a module you did not take, one more list follows — Recommended with your selection — add any?

Where should the module skills be installed? Into the project, or into your home directory for one copy across every app you scaffold. With both Claude Code and Codex CLI installed there is a question before this one: which of them finishes the setup.

Paste your Chatfuel token: Take one from the token page. The wizard checks it against the API before going any further — a rejected token is asked for again, three times — and writes it to .env with mode 0600. It never reaches the browser.

Which Chatfuel workspace should the app open? Asked only when the account has more than one; an account with none gets its first workspace created here. Not which bot: the app lists your workspaces and their bots itself at runtime, so a bot you create next month is in the picker without a re-run.

What is this app called? and then the logo. The name defaults to the workspace's title and is the browser tab, the top bar and the sign-in screen; the image — svg, png, jpg, webp, gif or ico — becomes the tab icon and the mark beside it, copied into public/. Enter keeps the shipped logo.svg, which is a real mark rather than a placeholder, so an unanswered run still looks finished.

Which browser origins may call the proxy? Only the app's own origin is the recommended answer and what a normal deployment needs. * is a local-development setting: a deployed proxy refuses to start with it.

Where should the app be created? ./chatfuel-app unless you say otherwise. The directory has to be empty or absent.

Then it copies the template, writes .env and .chatfuel/lock.json, installs the skills and installs the dependencies. What follows is offers, each a yes or no: deploy to Vercel, put it on GitHub, launch your coding agent in the new app, start the dev server in your browser. Decline every one and the app is still complete.

With admin there is one more question, the panel's password; with auth, the Supabase ones. Both are on their module pages: Admin and Accounts.

Start it

npm run dev

The app is on localhost:5173.

Check that it works

/ opens the first installed module, with the rest grouped in the sidebar beside it. With exactly one module installed there is no sidebar at all.

In the top bar, a picker listing every workspace the token's account owns and the bots inside each. That list is read live from the API, so seeing it is proof the token reached Chatfuel through the proxy. No bot is named anywhere in the code.

Click into a second module and watch the address bar. It reads /<moduleId>, and a module with more than one screen adds its own segment — /contacts/fields, /bookings/calendar?mode=week. Reload that address: these are real paths, not fragments, and the dev server answers an unknown one with index.html, so it comes back on the same screen.