Support, Chatbot & Knowledge Base
Answer the easy questions instantly. Hand over the rest cleanly.
A layered engine works from the cheapest, most predictable answer to the most expensive one — and every decision records which layer answered, so you can see exactly what the bot is doing and what it is guessing at.
Analytics
Updated just now
Deflected
84%
↑ 6pp
Conversations
1,908
Median reply
1.2s
Which layer answered
Top unanswered questions
Question to resolution
Six layers, tried in order, cheapest first.
The order matters more than the cleverness. A quick reply that always works beats a model that usually does.
0107Sayl
Business hours, opt-outs and safety checks
Before anything is answered, the guards decide whether it should be. Someone who has opted out is not messaged, and out-of-hours behaviour is your configuration rather than a default.
Happens on /chatbot in the product
In the product
What you get, and why it is there.
Configuration, not code
Over a hundred options across sixteen categories, rendered from a schema. A new option appears in the settings screen without a release.
Draft and published are separate
You edit a draft and publish it deliberately. Nobody has ever wanted to edit a live chatbot in place.
Version history with rollback
Restoring an old version loads it into the draft rather than pushing it straight to live, so a rollback is still a decision.
Self-hosted knowledge base
Your articles, on your domain, with CSV, JSON and Markdown upload. HTML is sanitised against an allowlist rather than a list of known-bad tags.
A diagnostics page
A health screen that tells you which parts of the engine are configured and working, so 'the bot is not replying' is a two-minute question.
Audited opt-out clearing
The opt-out list can be cleared, and clearing it is recorded. Consent state is not something that should be quietly editable.
Where each step actually happens
These are real screens in the product, not stages in a diagram. You can be looking at every one of them on day one.
See the whole platform mapGuards
Business hours, opt-outs and safety checks
Quick reply
The exact answers you wrote, matched exactly
FAQ
Curated question-and-answer pairs
Knowledge
Retrieval across your own documents
Handoff
A person takes over on the same thread
Measure
Every decision is recorded with the layer that made it
Straight answers
What this does not do
Every one of these is a real constraint — some ours, some the upstream platform's. They are on this page so you find them now rather than after you have paid.
History imports are deliberately excluded
When you connect a number, the bot does not reply to imported historical conversations. Without that rule, connecting a number would message every customer you have ever spoken to.
The AI layer needs a key
The retrieval and AI layers are built and tested, but they are inert until an API key is configured. Until then the engine runs deterministically, which is a real mode rather than a broken one.
Keep going
What connects to this
Ready when you are
See Support running on your own process.
Bring a real example — an enquiry that got lost, a quote that took three days, a campaign nobody could measure. We will walk it through the actual product.