GENChat: AI chatbots, governed

Your content

Answers come from what you have published: your website, knowledge base, FAQs and documentation, indexed so the most relevant passages are found for each question. What is not relevant enough is left out, not stretched to fit.

Live data

Stock, orders, bookings or service status, answered from the system that holds the facts, through a controlled integration layer.

Hand-over

When the model is unsure, the customer is frustrated, or they simply ask for a person, a member of your team joins the same conversation through a secure link, with the whole history in front of them. Out of hours it takes their details instead.

Models

Our own models run locally on GEN hardware, so conversations need not go to a cloud AI provider, or it can use backends of your choosing.

One tag

Added to any site with a single script tag. It follows your light or dark theme, and its position, greeting and appearance are yours to set, down to the CSS.

Sessions

The conversation carries on as the visitor moves between pages, and can pop out into its own window.

Security

It runs only on the domains you authorise, inside a sandboxed frame, with CORS controls.

Multi-tenant

Each site has its own configuration, policies, content and backends, isolated from the others.

Monitoring

Every conversation is kept with its full transcript, the pages the visitor viewed, IP address and reverse DNS, and user agent, for audit and for improving the answers.

A chatbot should never be connected directly to a language model. One left to its own devices answers from whatever it learned in training and fills the gaps with confident invention. Production assistants need governance: safety constraints, policy enforcement, tool routing, validation, caching, and deterministic fallbacks for the tasks that must always work.

Governance

A policy and routing layer controls behaviour, not the model.

Determinism

Reliable flows and fallbacks wherever a task has to work the same way every time.

Grounding

Responses built from trusted endpoints, not free-form generation.

Isolation

Separate configuration, domains and backends for every site.

Version 1

Pure logic and algorithms: deterministic routing and outcomes driven by explicit rules, state machines and classical inference.

Version 2

Language models added for inference signals, determining intent and extracting detail, while the platform stayed primarily logic driven.

Version 3

The first release in which finely tuned, tightly constrained models take part in real conversation, orchestrated by deterministic logic and grounded in several information endpoints. It runs on GAIN, our multi-inference AI platform.

The chat at the bottom right of this page is GENChat. Ask it about our software. To put one on your own site, tell us about the questions your customers ask most.