How to Add an AI Web Agent to Your Website (No Code)

Short answer: You can add an AI agent to your website with Matram without programming or building conversation flows. Create an agent, connect your website or documents, test its answers, configure the widget and paste the script tag from Matram’s Installation tab into your website’s custom-code area. The script is supplied for you; you copy it rather than write it.

With Matram, the real work is deciding what the agent should answer, giving it reliable content and testing its boundaries. Installation is one supplied script tag.

“No code” needs one honest clarification. You do not write JavaScript or build the agent from scratch, but somebody with access to your website must paste Matram’s tag into the site. On most content management systems and site builders, that is a custom-code field rather than a development task.

Here is the complete process.

Before you add an AI agent to your website

Start with one sentence that defines the agent’s job. For example:

Answer pre-sales and support questions using our product pages, pricing page, help centre and refund policy. Capture interested visitors and offer a person when the content does not contain the answer.

That sentence controls what you connect and test. Old pages, campaign copy and draft policies create more opportunities to retrieve the wrong information.

Prepare these items first:

• the current pages or documents that contain approved answers;
• 20–30 real questions from support emails, live chat or sales calls;
• a name and welcome message for the agent;
• the team member who should receive escalations; and
• administrator access to your website or its custom-code settings.

If you are still deciding what belongs inside the agent, Matram’s guide to chatbot use cases helps separate useful website conversations from tasks that require another system or a person.

Step 1: Create your Matram AI web agent

Create a Matram account and open Chatbots in the main navigation. Select the option to create a chatbot, give it a clear internal name and open the Agent tab.

Use a name that identifies the purpose or site. “Main website support” is easier to recognise than “New chatbot.”

You can also set the welcome message here. Keep it specific enough to show visitors what the agent can do:

Hi! Ask me about plans, features, setup or account support. I’ll answer from our official information and show you the source.

A focused opening sets an accurate expectation. Avoid claiming that it can “help with anything.”

Step 2: Train the agent on your website and documents

A new Matram agent has no business knowledge until you add sources. Open Website Links or Files & Data Sources and choose the source that contains the answers visitors need.

Matram supports three useful starting routes:

1. Website source: Add a single URL or let Matram crawl linked pages on the same domain. One website source can collect up to 60 pages.
2. Sitemap source: Add your sitemap when the site is larger. Matram can read up to 500 URLs from a sitemap source.
3. Files or text: Upload PDF, DOCX, TXT, Markdown or CSV files up to 10 MB each, or paste a text snippet for information that is not published elsewhere.

Do not upload the same policy in several formats unless every version agrees. Conflicting information weakens retrieval.

When a source shows as ready, its content is available. Large crawls can finish partially after reaching limits, so check which pages were collected.

For a more detailed content-planning process, use Matram’s guide on how to build a chatbot.

Step 3: Set the agent’s answer boundaries

Open the response settings and decide how closely Matram should stay within your connected material.

Strict mode is the stronger default for company-specific support. It answers from trained content and declines when that content does not support a response. Flexible mode permits small talk and general knowledge, while business facts still come from your sources.

Strict mode does not correct weak documentation. If a valid pricing question is declined, improve the source rather than instructing the agent to guess.

Use persona and instruction settings for tone, vocabulary and scope—not to override source facts. For example:

Keep answers concise. Use our product names exactly as written. Never invent a feature, price or policy. Offer a person when the source does not answer the question.

Step 4: Configure the widget without touching code

Open Customizations → Appearance. This is where you control what visitors see before the agent reaches the website.

Matram lets you set:

• widget colours;
• bottom-left or bottom-right position;
• agent avatar and name;
• launcher-icon size;
• chat height and font size;
• tooltip and input-placeholder text;
• auto-open behaviour for desktop and mobile;
• proactive messages;
• conversation starters and follow-up suggestions; and
• right-to-left layout for languages such as Arabic.

Match the site’s visual system, but prioritise readability. Use auto-open and proactive messages sparingly so the agent does not behave like a pop-up.

Conversation starters work best when they reflect real intent: “Which plan suits a five-person team?” is more useful than “Tell me more.”

Step 5: Set lead capture and human handoff

An agent should not trap a visitor when the question needs a person. Matram supports lead capture and human handoff inside the same widget.

The lead card collects a required name and email, with optional phone and location. It can appear after engagement, during escalation or under a lead-capture or demo-booking goal. Captured leads appear in the Leads tab with a summary and CSV export.

Visitors request a person in their own words, while clear complaints can also trigger escalation. Your team receives an email and replies from the Inbox or Chat History. The bot stays paused after takeover.

Configure and test this route before publishing. A good answer is valuable, but a clean escape route matters when the answer is unavailable, sensitive or account-specific.

Step 6: Test the AI agent before it reaches customers

Use the live test panel on the Agent tab or the full-page Playground. Both use the same sources and settings as the live agent.

Use real customer wording, including misspellings, incomplete questions and contextual follow-ups—not only polished marketing questions.

Your pre-launch set should include:

• common product, pricing, delivery or policy questions;
• questions that require information from two different pages;
• a follow-up that depends on the previous answer;
• one question the source material definitely does not answer;
• a direct request for a person; and
• a complaint that should escalate.

Check the source pills, not only the wording. Matram can show up to four source citations below an answer. Open them and confirm that the cited page genuinely supports the response.

When an answer is missing, fix the content: crawl the relevant page, add the document, write a snippet or use a fixed custom response for exact wording. Matram’s chatbot best-practices guide explains how to test historical conversations and build a content backlog.

Step 7: Add the Matram AI agent to your website

Once the answers and handoff hold up, open the agent’s Installation tab. Matram displays a single script tag containing the public bot ID.

To install it:

1. Copy the complete script tag from Matram.
2. Open your website’s HTML or custom-code settings.
3. Paste the tag immediately before the closing </body> tag on every page where the agent should appear.
4. Save or publish the website changes.
5. Reload a live page and open the chat bubble.

That tag loads the configured appearance, citations, starters, proactive messages and lead capture. It contains a public bot identifier, not a private API key.

The widget works wherever custom script tags are allowed. If your website plan blocks them, ask the administrator to add the tag or enable custom code.

Matram also offers a REST endpoint for custom interfaces, but that is a developer route. Standard websites should use the supplied widget.

Step 8: Verify the live installation

Do not stop when the bubble appears. Run a short live-site check on desktop and mobile:

• open and close the widget;
• send one documented question and open its citation;
• send one unsupported question and confirm the response is controlled;
• submit the lead form and check the Leads tab;
• request a person and verify that the notification arrives;
• confirm that the widget does not cover navigation, cookie controls or checkout buttons; and
• test the pages where the widget should and should not appear.

The widget uses Shadow DOM so its styles and the site’s CSS do not interfere. Dashboard changes can take about 60 seconds to appear, so wait and refresh before troubleshooting.

What to review after launch

Read the first conversations rather than assuming installation equals success. Look for unanswered questions, weak citations, repeated requests for the same missing detail and moments where visitors expected a person sooner.

Update sources when the business changes. Matram resyncs website and sitemap sources by plan schedule, and you can refresh them manually.

A simple weekly routine is enough at the start: review conversations, fix the highest-volume content gap, retest it and check that handoffs reached the right person.

Frequently asked questions

Can I add an AI agent to my website without coding?

Yes. Matram handles agent configuration, training, testing and widget behaviour in its dashboard. Installation involves copying its supplied script tag into your website’s custom-code area; you do not write the code yourself.

Can I train Matram on my entire website?

One website source can crawl up to 60 pages, while a sitemap can supply up to 500 URLs. Larger libraries can be divided across sources, subject to plan limits.

How do I know whether the agent is answering correctly?

Test real questions in the Playground and open each citation. Also test an unsupported question to confirm the agent declines or hands off.

Will the AI agent match my website design?

You can set colours, position, avatar, launcher size, font size, chat height, tooltip and other behaviour from Matram’s Appearance settings. The widget is isolated from the site’s CSS through Shadow DOM.

Add your first AI agent with Matram

Spend more time on sources and testing than on the embed. Matram installation is one copied tag; quality comes from current content, clear boundaries, verified citations and human handoff.

Connect only trusted information and test difficult questions, not just easy demo prompts. Once the answers hold up, add the script and repeat the tests live.

You can create a Matram agent and train it on your own site before publishing the widget.