Cephei Knowledge Assistant

An AI assistant for your website.

CKA generalizes our AI Chat Prompt into an embeddable website assistant. The widget answers questions based on your website context, your configuration, and clear safety boundaries. CKA is currently in pilot phase; there is no general self-service purchase yet.

Origin From the Cephei AI Chat Prompt to a website widget Our chat prompt demo shows the technical foundation: local AI backend, streaming, website context, and clear limits. CKA turns that into a reusable widget for other websites. View AI Chat Prompt
Embeddable one script tag integrates the assistant into the website
Tenant-based context, theme, limits, and allowed domains per website
Pilot phase open for suitable websites, but not a standard product yet

Currently a pilot programme, not a shop product

CKA is functional and already running in real integrations. At the same time, the pilot phase still evaluates onboarding, context maintenance, operations, privacy wording, and suitable industry scenarios. If you want to participate, contact us; we will assess together whether the website and use case fit.

Open pilot programme Individual setup No self-service booking

What CKA is built for

Many websites already contain the relevant information, but visitors do not always find it quickly or ask questions that menus and search boxes handle poorly.

Website knowledge is distributed

Services, process information, blog posts, or FAQs live across multiple pages. CKA can turn them into a structured answer base for visitor questions.

Visitors do not search like editors think

People ask in everyday language. A website assistant can receive these questions and answer them from curated content.

Sensitive contexts need boundaries

Medical practices, advisory services, and company sites need clear notices, limits, prompt-injection protection, and controlled answer logic.

A chatbot alone is not enough

The decisive pieces are tenant configuration, context maintenance, allowed origins, privacy link, limits, monitoring, and a realistic operating model.

What is inside the CKA system

The widget is only the visible surface. The product logic lives in the multi-tenant backend and in the controlled configuration for each website.

Module 1

Knowledge packs

Curated website contexts, article indexes, or practice information form the answer base for each tenant.

Module 2

Local AI backend

The model integration runs server-side through Cephei infrastructure. API keys and model access stay out of the browser.

Module 3

Safety boundaries

CSRF, origin allowlisting, rate limits, prompt limits, prompt-injection detection, and an optional AI safeguard reduce typical risks.

Module 4

Web search and sources

Depending on the tenant, web search can be enabled. Answers can include source references when the configuration provides for it.

Module 5

WordPress pilot integration

Small tenant-specific WordPress plugins are currently in use. A generic plugin and self-service onboarding are later product phases.

Live pilots

CKA is already live

CKA runs, for example, on our own blog and on an ENT practice website. These examples show different contexts: editorial content on cephei.blog and curated practice information on hno-augsburg.com.

How a pilot works

A CKA pilot is not just a script installation. We first review the goal, content, boundaries, and operations.

Step 1

Review website and use case

Which questions should be answered, which content is suitable, and which topics must deliberately be excluded?

Step 2

Configure context and tenant

We create the initial answer base and define theme, limits, allowed domains, privacy link, and behaviour.

Step 3

Integrate and test the widget

The website loads the central widget. We then test common questions, failure states, and rendering on desktop and mobile.

Step 4

Evaluate the pilot

We assess value, answer quality, maintenance effort, and operating risks before turning this into a lasting offer.

What a pilot should clarify

  • which questions visitors actually ask
  • whether the available website context is sufficient
  • which boundaries and notices are necessary
  • which operating and maintenance model is viable
Request a pilot

Context for sensitive websites

CKA does not replace professional advice or legal, medical, or security-critical decisions. The assistant provides orientation based on the supplied context and points to appropriate contact paths when needed.

Cookieless widget model Tenant isolation Clear answer boundaries

Why Cephei

Interested in the CKA pilot programme?

If you want to test a website assistant for your blog, practice, company website, or another suitable scenario, write to us. We will get back to you for an initial assessment.

Common questions about the CKA pilot

Before a pilot, it is important to understand what CKA can do, where the boundaries are, and how integration works.

Can CKA simply be purchased today?

No. CKA is currently in pilot and evaluation phase. The pilot programme is open, but every participation is assessed and set up individually.

Which websites are suitable for a pilot?

Suitable websites have clear, curatable content: blogs, practices, company sites, product or service information. The important part is that intended answers can be grounded in reliable sources.

What needs to be integrated technically?

At the core, the website loads a central JavaScript widget with a tenant ID. Security-relevant configuration, context, and model integration live server-side in CKA.

Does the widget set cookies?

The current CKA widget is designed to be cookieless. Session and CSRF data are held in widget storage. Independently of that, the website must describe AI processing properly in its privacy policy.

Can CKA be operated in a GDPR-compliant way?

CKA is designed for privacy-conscious operation. Processing runs on Cephei-owned hardware in our own rooms at a German location, not on third-party cloud infrastructure. For each pilot, we also clarify the appropriate privacy notice wording and organizational agreements.

Are inputs used for training or shared?

No. Inputs and answers are not used for AI training and are not disclosed to third parties for their use. Conversation content is not stored permanently; technical processing is limited to answering the request and limiting misuse.