Silence beats a wrong answer.

OmniClew exists because the people who ask a website questions deserve an answer they can trust, and the organisation behind the website deserves to know what its website is saying.

Visitors ask in their own words

People ask websites questions in their own words, and websites answer in official language, spread across pages and PDFs. A visitor asking “can I put a grill on my balcony?” will not find a section called “Open flame appliances”. Search returns a list of pages, and the visitor has to read them to find the answer. Many visitors give up and send an email instead.

A chatbot at runtime is a gamble

An AI chatbot generates a new answer every time, so every answer is a chance to be wrong. A chatbot that gets a fee, a deadline or a rule wrong creates a support problem and can create liability for the organisation. A chatbot also costs money for each question, runs on someone else’s servers, and cannot be fully checked before it speaks.

Do the thinking before deployment

OmniClew’s position is that the thinking should happen before deployment, and the visitor should only ever see something that was checked. OmniClew uses a large language model at build time, where a person can review its work. It then publishes only answers that are verified, quoted from the source, and calibrated. At runtime there is nothing left to hallucinate.

Prefer silence to a guess

A wrong answer from a trusted source does more damage than no answer. When OmniClew does not know, it says nothing, or shows documents that may help. It never composes text, and it never fills a gap with something plausible.

Find where your website disagrees with itself

A chatbot reading a website that contradicts itself will be wrong some of the time, however good the model is. A build compares answers across the whole website and flags the places where the website disagrees with itself, such as two different fees for the same thing. The organisation fixes the source, and every answer improves.

What OmniClew will not do

OmniClew will not answer questions that were never anticipated by the content. It does not reason across several documents to build a new answer, and it does not hold a conversation. A runtime language model is better at those tasks. OmniClew is built for organisations that value a right answer to the common questions over an answer to every question.

The OmniClew website is built to be its own first customer. If the omnibox is good enough to run a customer’s documentation, it is good enough to run this site. Every page here is reachable by typing what you want.