From signup to live chat in 10 minutes (the exact path)

The shape of it

Four steps. The whole thing should take ten minutes if you have your Gemini or OpenAI API key handy and one help-center article ready to paste. If you want it to take twenty minutes because you're also customising brand colours, fine. The widget is live the moment you paste the snippet.

1. Sign up: create the workspace.
2. Plug in your LLM: one API key, one dropdown.
3. Drop in one document: PDF, URL, or paste.
4. Paste the snippet: one <script> tag.

Step 1: Sign up (60 seconds)

Go to app.openagent.in/admin/signup. Email, password, workspace name. The workspace name becomes part of the widget's default identity (you can change it later under Settings → Workspace).

We approve new workspaces inside an hour during US business hours; evenings and weekends we're slightly slower. You get an email the moment your workspace is live with a sign-in link.

Step 2: Plug in your LLM (90 seconds)

Once you're signed in, head to Settings → Providers and click Add provider. Pick from Gemini, OpenAI, Anthropic, OpenRouter, or “OpenAI-compatible endpoint” (for Ollama, vLLM, TGI, LM Studio, Together, Groq).

Paste the API key. The platform encrypts it on submit. If you're new to LLM providers, our BYO LLM guide covers what each provider does well.

Step 3: Drop in one document (3-5 minutes)

Open Knowledge → Add document. Three input modes: upload PDF/DOCX/MD/TXT, paste a URL we'll fetch, or paste raw text. Pick whichever maps to where your help content already lives.

For most workspaces, the fastest path is pasting a help-center URL. The platform fetches, strips navigation/footer boilerplate, chunks, embeds, and indexes. Indexing is asynchronous; small documents are searchable within 10-30 seconds, longer documents finish in 1-2 minutes.

Don't overthink this step. One document is enough to demonstrate the AI answering from your data. You can come back and add 50 more later. The Pro plan ships 5 GB of attachment storage per site, which is roughly “all your help-center articles plus the entire PDF knowledge base” for most workspaces.

Background: read how RAG works if you want to know what's happening on the server while you wait for the green tick.

Step 4: Paste the snippet (60 seconds)

Open Settings → Install widget. Copy the one-line snippet. Paste it right before </body> on the site you want the widget on.

<script>
  window.OPENAGENT_CONFIG = {
    tenantSlug: "yoursite",
    apiUrl: "https://app.openagent.in",
    wsUrl: "wss://app.openagent.in"
  };
</script>
<script async src="https://app.openagent.in/openagent.js"></script>

The widget loads lazily after the page's first paint, renders inside a Shadow DOM so it doesn't collide with your site's CSS, and the bubble appears bottom-right. Click it, ask the AI something covered by your knowledge doc, you're done.

What comes after

Once the widget is live, the next four things teams usually do (in rough order of impact):

1. Add more knowledge documents. One doc is enough to prove the path; ten docs makes the AI useful across most of your queue. Upload the help center, the refund policy, the shipping policy, the FAQ.
2. Brand the widget. Logo, accent colour, bubble position. See the branding guide for what visitors actually evaluate.
3. Wire alerts to Slack or Google Chat. Settings → Alerts, paste a webhook URL, get pinged the second a visitor types. So you don't have to live in the dashboard.
4. Tune the system prompt. Agents → Edit. The default prompt is fine; the second-revision version that mentions your tone of voice and refusal style is better.

None of these are blocking. The widget works the moment you paste the snippet. Everything after is iteration.