How to Deploy an AI Chatbot on Your Website with Macha (Step by Step)
Putting an AI chatbot on your website used to mean a flow-builder, a pile of intents, and a week of work. With Macha it's a few minutes: you pick one of your agents, style the widget, and paste a single line of code. Because the chatbot is powered by a real agent, it doesn't just answer FAQs from a script — it uses your knowledge and can take actions, then hands off to a human when it needs to.
This guide walks through deploying one end to end.
What a Macha chatbot actually is
A Macha chatbot is a public, customer-facing widget you embed on your website or help center. The important part: it's powered by one of your agents, so it inherits that agent's instructions, knowledge sources, and tools. Whatever the agent can do in Macha, the chatbot can do for your visitors — answer from your help center, look up an order, check a subscription — grounded in your real content rather than a generic model.
That's the difference between a scripted website bot and this: a scripted bot follows a decision tree; a Macha chatbot reasons over your knowledge and acts. (For the deflect-vs-resolve angle specifically, see how to add a Zendesk AI chatbot that resolves.)
Before you start
- An agent to power it. If you haven't built one yet, do that first — give it the knowledge and tools you want the chatbot to use. (See what a Macha agent is.) A "Support" agent grounded in your help center is the usual choice.
- The right plan. Website chatbots are available on Professional ($699/mo, up to 3 chatbots) and Enterprise (up to 20).
- Access to your site's HTML (or your CMS/site builder) to paste one line of code.
The three-step model
Everything below fits in three moves: pick an agent → configure the widget → embed it. The Chatbots page lays it out exactly this way.
Step 1 — Create a chatbot and pick its agent
Go to Chatbots and click Create Chatbot. Give it a name (internal — your visitors won't see it) and choose the agent that will power it. This is the single most important choice: the agent's knowledge and tools are the chatbot's brain. Point it at a support agent grounded in your help center and the chatbot answers from your real content.
Step 2 — Configure the widget
This is where you make it look and behave the way you want. As you change settings, the live preview on the right updates so you can see exactly what visitors will get.
- Header title — the text at the top of the chat panel (e.g., "Chat with us").
- Greeting — the first message visitors see (e.g., "Hi! What can I help with today?").
- Avatar — upload an image to brand the bot.
- Primary colour — match your brand; it tints the header and launcher.
- Position — anchor the launcher to a corner. You set Top / Right / Bottom / Left independently (in
px,em, orrem) — fill the two sides you want and leave the opposite pair blank. Bottom-right is the convention. - Allowed domains — restrict the widget to load only on domains you own, so nobody can lift your embed code onto another site.
- Tool safety (read-only vs. write) — decide whether the public chatbot can only read (search knowledge, look things up) or also run write actions. On a public widget, write actions execute without a confirmation step, so scope this deliberately — start read-only and open up specific write tools only when you trust the setup.
Step 3 — Embed it on your site
When you're happy, Macha gives you a single line of code — a <script> tag tied to this chatbot. Paste it into your site's <head> (or your CMS's "custom code"/"footer scripts" area), publish, and a floating launcher appears in the corner.
That's the whole install. A few things happen automatically once it's live:
- Visitor sessions persist across page loads (via
localStorage), so a returning visitor keeps their conversation. - Every conversation shows up in your Macha dashboard, so you can review what visitors asked and how the bot answered.
Step 4 — Test it before you point traffic at it
Open your site, click the launcher, and run it through the questions you actually get:
- Ask a common FAQ — does it answer from your help center accurately?
- Ask something it shouldn't answer — does it decline or escalate cleanly instead of guessing?
- If you enabled an action (like an order lookup), test it with real-ish data.
Watch the first batch of real conversations in the dashboard and tighten the agent's instructions or knowledge based on what you see. The chatbot is only as good as the agent behind it, so most tuning happens on the agent, not the widget.
Customization recap
| Setting | What it controls |
|---|---|
| Agent | The chatbot's knowledge, tools, and behavior |
| Header title & greeting | First impression / framing |
| Avatar & primary colour | Branding |
| Position | Which corner the launcher sits in |
| Allowed domains | Where the widget is permitted to load |
| Read-only vs. write | What the public bot is allowed to do |
Where to put your chatbot (and which agent to use)
The same widget behaves very differently depending on where it lives — match the agent and the placement to the page:
- Help center / docs — a support agent grounded in your articles. This is the highest-value spot: visitors are already trying to self-serve, and the bot can resolve them before they ever file a ticket.
- Homepage & product pages — a pre-sales agent that answers "does it do X?" and points to the right page. Catches buyers before they bounce.
- Pricing / checkout — answers plan and billing questions at the moment of highest intent, where a stuck visitor often means a lost sale.
- Inside your app — a support agent scoped to account-aware tools, for logged-in users.
You can run a different chatbot (and a different agent) on each surface, up to your plan's limit — so the docs bot, the pre-sales bot, and the in-app bot can each be tuned for their job.
A support chatbot in under 10 minutes — a worked example
Say you want a help-center bot that deflects common questions:
- Build a Support agent. Instructions: "Answer support questions from the help center. If you're not confident, or the request needs account access, tell the visitor you'll connect them to the team." Knowledge: your Zendesk Help Center (auto-synced). Tools: read-only.
- Create a chatbot and pick that agent.
- Style it — header "Need help?", greeting "Ask me anything about [product]", your brand colour, anchored bottom-right.
- Lock allowed domains to your docs domain only.
- Embed the one-line script in your help center's
<head>and publish. - Test five real questions, then watch the first transcripts and refine the agent's instructions.
Ten minutes of work, and self-serve questions get answered instantly instead of turning into tickets — and the ones the bot can't handle escalate cleanly with context.
Best practices
- Ground the agent well. Keep its help-center/knowledge current — that's where answer quality comes from.
- Start read-only. Add write actions only once you trust the answers.
- Set a clear greeting that tells visitors what the bot can help with, and make escalation to a human easy.
- Lock the allowed domains so your embed can't be reused elsewhere.
- Review transcripts weekly for the first month and refine the agent.
Frequently asked questions
How long does it take to deploy a Macha chatbot? Minutes. Pick an agent, configure the widget, and paste one line of code.
What powers the chatbot's answers? An agent you choose — it uses that agent's knowledge sources and tools, so answers come from your real content, not a generic model.
Where do I paste the embed code? Into your site's <head>, or the "custom code"/"footer scripts" section of your CMS or site builder.
Can I limit where the chatbot appears? Yes — set allowed domains so the widget only loads on sites you own.
Can the chatbot take actions, not just answer? Yes, if you enable write tools — but on a public widget they run without a confirmation step, so scope them carefully and start read-only.
Which plans include chatbots? Professional (up to 3) and Enterprise (up to 20).
The bottom line
A Macha chatbot turns one of your agents into a website assistant in three steps — pick the agent, style the widget, paste the code. Because it's agent-powered, it answers from your real knowledge and can act, not just deflect. Build the agent well, start it read-only, watch the transcripts, and you've got a resolving chatbot live on your site in an afternoon.
Deploy your first chatbot: pick an agent, style it, and embed it in minutes. 7-day free trial, no credit card required. Start free.
