Introduction

Get Started

Five steps from an empty account to an assistant answering questions on your site, using only content you have provided. No prompt writing and no model configuration.

Create an agent

An agent is the assistant your visitors talk to. Give it a name, an optional description, and choose whether the widget renders light or dark. Nothing else is required before it will answer.

You can create your first agent on the Agents(sign in required) page.

Create a knowledge base

A knowledge base groups the material an agent may use when answering. When you create one you choose its scope, either available to every agent on the account or only to agents you name. Use scoping when different assistants should not share the same material.

Add your sources

There are three ways to put content into a knowledge base:

Question and answer pairs, entered one at a time or as a batch. These are matched directly and are the most precise option for policies and common questions.
Documents, as PDF or DOCX, up to 25 MB each. Text is extracted and indexed automatically, so scanned pages without a text layer will not work.
Plain text, pasted straight in with an optional title.

Indexing runs in the background. A source moves from processing to active on its own, and the agent begins using it as soon as it does.

Test it privately

Open the Playground(sign in required) and ask the questions your visitors actually ask. Answers are drawn only from your sources. If the material does not cover a question the assistant will say so rather than inventing an answer, which is the behaviour to check for before going live.

Deploy it to your visitors

Collect your credentials from the API Keys(sign in required) page. Create a secret key beginning wcx_sk_ for server calls, or create a widget installation to be issued a publishable key beginning wcx_pk_ together with its embed snippet.

For a website, add your origin to the installation's allowed origins, then add one line before your closing body tag:

index.html

<script
  src="https://api.wicarax.app/v1/widget.js"
  data-wicarax-key="wcx_pk_YOUR_PUBLISHABLE_KEY"
  async
></script>

To build your own interface, call the API from your server with the secret key instead:

terminal

curl https://api.wicarax.app/v1/chat/completions \
  -H "Authorization: Bearer $WICARAX_SECRET_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "wicarax-zenith",
    "messages": [{ "role": "user", "content": "Your question" }]
  }'

A publishable key is public by design and only works from an origin registered on its installation, so an installation that allows no origin answers nowhere. A secret key is not origin checked and must stay on your server. Responses are returned complete in a single body, so there is no partial delivery to handle.

Choose an integration

Use	Credential	Where it runs
Hosted widget	Publishable key from a widget installation	A browser on an allowed origin
Official SDK chat	Secret key	Your server
REST API chat	Secret key	Your server or backend automation

The production API URL shown in examples is the intended deployment address. Use your configured API origin for the environment you are testing; local development may use http://localhost:4000.

Before you launch

Check	Expected result
Supported question	A specific answer grounded in active material
Unsupported question	A clear decline without invented facts
Website origin	An exact origin configured on the widget installation
Credential location	Publishable in the page; secret only on the server
Usage	Enough monthly quota and per-minute capacity for the expected traffic

Tip

Seed your first knowledge base from your support inbox rather than from an internal brainstorm. Ten questions customers actually sent, in their words, will outperform fifty you imagined.

Disclaimer

An assistant answers from the material you provide and nothing else. It has no access to your website, your database or the wider internet unless you add that content as a source, and when your material does not cover a question it says so rather than guessing. Accuracy is therefore a property of your sources, which is why step four exists.