Live in 5 minutes

By the end of this guide you'll have a widget answering from your knowledge base on your site, and an agent that has completed its first task. You only need to touch code at the embed step (copy-paste).

What is SemAgent?

SemAgent is a single platform with two products:

• RAG Widget — a chat widget that answers visitors with citations, from a knowledge base built from your site and documents.
• AI Agent — an agent that gets work done on your behalf (and with your approval) across email, calendar, Drive, databases and Shopify.

Both share the same knowledge base: everything you teach the widget, the agent knows too.

Quickstart: set up the Widget (5 min)

1. Create an account and start the knowledge base

1. Start your free trial at semagent.ai/register (7 days, 250 credits — no credit card).
2. In the panel, go to Knowledge Base → Add Source and enter your site URL.
3. Click Start Crawl. Your sitemap is found automatically; if you have a forum, its engine (Discourse, phpBB, XenForo, vBulletin, inSided) is auto-detected.

2. Add the embed code to your site

Get your tenant ID from Widget → Setup in the panel and paste this one line just before your </body> tag:

<script src="https://semagent.ai/static/widget.js"
        data-tenant-id="your-slug" defer></script>

// app/layout.tsx
import Script from "next/script";

<Script src="https://semagent.ai/static/widget.js"
        data-tenant-id="your-slug"
        strategy="lazyOnload" />

// Appearance → Theme Editor → footer.php
// (or a "header/footer scripts" plugin)
<script src="https://semagent.ai/static/widget.js"
        data-tenant-id="your-slug" defer></script>

3. Test it

1. Refresh your site — the widget bubble appears at the bottom right.
2. Ask a question your site does answer → you get a cited answer.
3. Ask a question your site doesn't answer → the widget says it doesn't know instead of making something up. That's strict-KB grounding working.

Quickstart: your first agent task

The agent works by chatting in the Agent tab — no separate setup. Your knowledge base is enough for a first task; connectors are optional.

1. Choose your autonomy level

Agent → Settings → Autonomy: if you're just starting, stay on Supervised — reading and drafting run automatically, sending/writing actions come to you for approval.

2. Give the first task

Prepare a short, customer-ready document summarizing
our return and shipping policy from the knowledge base.

The agent searches the knowledge base with kb_search and produces a Word document with doc_generate. Both are read/draft (T1) tools, so no approval is needed — the result lands as an artifact in a few minutes.

3. Approve an action

Email this document to me.

Sending email is a T2 (action) tool: the agent prepares a draft and shows an approval card. When you click "Approve", it sends. You've just seen the approval flow — the heart of the agent's safety model.

4. (Optional) Schedule the task

Every Monday at 09:00, summarize new content added to
the knowledge base and email it to me.

The agent schedules it with schedule_task; it appears under Settings → Scheduled Tasks and can be cancelled from chat.

Core concepts

Next steps

• Build your knowledge base — crawl sources, forums, files, the quality loop.
• Agent tool reference — all 50+ tools, risk classes and approval behavior.
• Widget Chat API — for building your own interface.