Quickstart

From signup to a live agent answering questions on your site in under five minutes. Each step is sized so you can stop and pick it up later.

1. Create an account

Visit the home page and click Get started. Blengi uses Laravel Fortify for auth, so you'll get standard email + password registration, plus optional 2FA in your account settings later.

On signup we create your first workspace. If you entered your website during signup, we also create your first AI employee for it: named after your domain, the domain already allowed, and reading your homepage straight away. Without a website you land on Home, which asks for one: Let us build your AI employee.

2. Run the onboarding wizard

The wizard at /onboarding is the one way to create an AI employee. Home's website field opens it and starts reading the site straight away; New AI employee on the AI employees page creates a new one and opens the wizard for it (the plan's AI employee limit applies). The wizard walks you through:

1. Confirm your website URL. The agent will learn from this site.
2. Auto-discover crawlable pages. We probe your sitemap and a handful of common paths (/about, /pricing, /faq, /docs, etc.).
3. Pick which pages to ingest. Each page becomes a knowledge source — text gets chunked and embedded into the vector store.
4. Watch them flip to "indexed". Polls every few seconds. Most sites finish in under a minute.
5. Copy the embed snippet. One line of HTML.

The last screen opens a live chat with the new AI employee (a test chat, never counted as a visitor). Done: go to Home takes you to Home, where the setup checklist shows what is left.

3. Embed the widget

Paste this just before </body> on any page where you want the agent to appear. Replace YOUR_AGENT_ID with the value the onboarding wizard shows you. The ready-made snippet, with the ID filled in, is also on the AI employee's Website tab under Install code.

<script
    src="https://your-app.test/widget/widget.js"
    data-agent-id="YOUR_AGENT_ID"
    async></script>

4. Publish the agent

A draft agent does not start on any website: the widget's /widget/init call answers 404 for it. Only the preview inside the dashboard can talk to a draft. Click Publish at the top of the AI employee's pages (it also publishes itself when its first source has been read). From then on the widget reads the agent's current settings, so every save is live for visitors straight away. Each publish also keeps a copy of the settings as an agent_version row, which Advanced › Earlier versions can restore; the widget never reads those copies.

5. Watch Home

Home lists what needs you: a visitor waiting for a person, new leads, questions your AI employee could not answer. Conversations appear live under Conversations, where you can read transcripts and take a chat over; captured leads are under Leads (/app/leads).

What's next?

Choosing which pages the agent learns from

On the first setup screen, under the website address, you pick how the starting knowledge is gathered:

• Show me what you found — the default. We scan, then you tick the pages before anything is added.
• Let Blengi choose — we scan and add the most useful pages straight away, so you reach a working agent without a decision. Nothing is hidden: the pages appear in Sources, and you can add or remove any of them afterwards.
• I'll enter them myself — skip the scan and paste the exact addresses, one per line. Duplicates are ignored and a bare example.com/about is understood as https://.

What "scan" means, precisely: we read your sitemap.xml (including any listed in robots.txt) and, when there is no sitemap, the links on your homepage. We do not walk the whole site. Anything missed can be added later, and pages a visitor opens are picked up automatically when that setting is on.

A page that cannot be fetched — a typo, a redirect loop, a login wall — is recorded as a failed source and skipped. The rest of the batch is added regardless. A source that failed for a passing reason (a timeout, a rate limit, a server error) is retried by a sweep every 15 minutes, up to three times; a typo, a missing page or a login wall is not retried automatically.

Talking to your agent before you install it

You do not have to put the snippet on your site to find out whether the agent works. As soon as the first page finishes indexing, the last setup screen shows your real widget and invites you to ask it something — while the remaining pages are still being read in the background.

It is the actual widget, not a demonstration: the same code, the same look, answering from the pages it has read so far. Nothing you ask it there counts as a conversation or a content gap. A lead form you send from it does create a real lead: it lands in Leads, so you can check the form and the email, but it is not counted in Results.

The install snippet sits underneath. Copy it when you are happy with what you see, rather than pasting code and hoping.