We built demo sites for invented companies and wired real Fixorix support into them. Open the chat and ask anything — it answers from that company's knowledge base, not in generalities.
A visitor asks about ingredients and allergens straight from the menu page — and gets an answer from that same menu, without waiting for an operator.
Every example is its own site, with its own look, its own menu and its own knowledge base.
A visitor asks about ingredients and allergens straight from the menu page — and gets an answer from that same menu, without waiting for an operator.
Things to ask the chat
A shopper checks an order status and the scenario calls the shop API to answer with specifics. Unhappy conversations are flagged as urgent by the AI, and a person takes it from there.
Things to ask the chat
A student asks about the timetable, course access and payment on Telegram and VK. An agent handles it and hands the hard cases to a tutor.
Things to ask the chat
A customer asks what is happening with their repair; the scenario looks it up and answers with the stage, the estimate and the date. Disputed cases become tickets, opened by a person.
Things to ask the chat
Users write in English on Discord while the knowledge base is Russian: the scenario translates the question, finds the answer and replies in the language it was asked in. Plans, limits and what counts as an event come from the same place.
Things to ask the chat
Four channels, one queue. Operators answer from a single window instead of switching between apps.
A bot behind a @BotFather token. Text, photos, files, video, buttons, location and contact.
Community messages: text, photos and documents, text buttons.
A bot on your server. Picks up any attachment — images, video, audio, archives.
Chat for your site: Markdown, files up to 10 MB, buttons under a message, recognises a signed-in visitor.
Everything on the demo sites is assembled from these parts, with no code written for any particular example.
A graph of blocks instead of code. Triggered by an event, a schedule, a manual run or an external system; branching, HTTP requests, tags and statuses. Every scenario has a draft, versions and a full run trace — which blocks it went through and what the variables held.
Reads the mood and urgency of a conversation, summarises a long thread, drafts a reply from your knowledge base and suggests a matching template. An assistant to the operator, not a replacement.
Upload PDF, DOCX, TXT or Markdown and the AI answers from them. Retrieval is hybrid: by meaning and by exact words with Russian morphology. The base is internal — customers never see it.
Twelve blocks: generation, classification, data extraction, translation, moderation, answering from the knowledge base. The important ones carry a confidence threshold and a separate "not sure" branch, so instead of a wrong answer the question goes to a human.
Conversations are tagged and routed to teams with their own hierarchy and visibility. Reply templates, and one customer profile across every channel.
A public REST API with keys and fine-grained scopes. A dashboard builder for your own chat and ticket metrics.
An ordinary scenario is a chain of steps drawn in advance. An agent works differently: you hand it the conversation and it decides what to do next.
Available in support scenarios.
Nothing here was written specially for the demo — it is all standard product functionality.
Each example gets its own Fixorix project. Inside it, a Widget channel is created and the site's domain is added to the allow-list. One script tag goes on the pages.
Everything the company knows about itself — menu, prices, composition, hours, delivery rules — is uploaded into the project's knowledge base. It is internal: customers never see it, the AI uses it.
A scenario greets the customer, asks the follow-up questions, searches the knowledge base and decides what happens next. It runs on an event, a schedule, a manual trigger or a call from an external system, and every run leaves a trace of the blocks it went through.
When a visitor writes in, Fixorix searches the knowledge base with hybrid retrieval and drafts an answer. If nothing relevant is found, confidence is lowered and the conversation goes to an operator.
Conversations get tags — composition, delivery, hours, lost property — and go to the right team. Teams are hierarchical and see only their own chats, while the customer profile stays one across every channel.
The menu on the site and the knowledge-base articles are built from a single file. That is why the price in the chat always matches the price on the page — by construction, not by luck.
The platform is real. The widget, the channels, the knowledge base, retrieval, scenarios and AI answers all behave exactly as they do for customers. Open the chat on any demo site and you are talking to production Fixorix.
The companies are invented. None of them exists: the names, addresses, phone numbers, prices and menus were made up for the demonstration. Nothing can actually be ordered.
Conversations from these demos land in real support projects and are read by people. Do not leave personal data here.
Create a project, upload what you already explain to customers every day, and put the widget on your site. Support starts answering in your own words.
