DockAI · Capabilities
The whole of it.
An AI assistant that answers customers from your own information, admits when it doesn't know, and hands over to a person with the whole conversation intact.
Widget builder
Change the brand, the tone, and the greeting — the preview updates live. Read to the end of the conversation: the last turn is the assistant refusing a question it has no grounding for, which is the behaviour worth evaluating.
Preview
Hi — ask me anything about your order.
What's your returns policy?
Returns are accepted within 30 days of delivery, unopened. You can start one from your order page.
↳ returns-policy.pdf · §2.1
Can you tell me why my delivery driver was late?
I don't have anything on that, so I'd rather not guess. I'll pass you to someone who does.
↳ no grounding · handed to a person
Every answer above points at where it came from. The last one points at nothing — so it does not answer. That refusal is a successful outcome, and it is also the most useful thing the system produces: it is a question your documentation cannot answer, reported by the customer who asked it.
Demonstration · Fixed sample data · No network — nothing here leaves your browser
Every capability
Grouped the way the platform is built.
Not a feature matrix with ticks in it. Each of these is a thing the system does, described in enough detail that you could tell whether it is what you need.
Building the assistant
Widget builder
Visual customisation — brand colour, tone, greeting, placement — with a live preview against a mock site.
One-line embed
A single line added to your site. No rebuild, no technical requirements on your side.
Tone control
How the assistant speaks, set by the people who own the brand voice rather than by a technical setting.
Knowledge and grounding
Knowledge ingestion
Load in your documents, your website, your FAQs, and your systems. Answers are drawn from your material, not from the AI's general memory of the internet.
Answers only from your material
No supporting passage, no answer. The assistant says it does not know and offers a person. This is built into how the system works, not a polite request to the AI.
Guardrails
The topics it will not go near and the claims it will not make, enforced by the system rather than left to the AI's judgement.
Doing things
Taking action
What the assistant may do on its own, and what needs a person, is decided by you before anything runs unattended.
Human handover
A person takes over with the full conversation, the material the assistant used, and what it already tried.
Multi-channel deployment
The same assistant, answering the same way, wherever your customers talk to you.
Knowing how it is doing
Conversation analytics
What people actually ask, as opposed to what you expected them to ask.
Gap detection
The questions the assistant had no material for. Each one is either a page to write or a decision not to cover that topic.
Unanswered-question reports
Every 'I don't know', collected. Each one is a good outcome and a useful piece of information — both at once.
We build systems like this for other organizations.
DockAI exists because we met the same operational failure enough times to engineer it properly. Yours is probably a different failure. That is the conversation we want, and you will have it with an engineer.