Write how it should sound.
Describe how the agent should talk, what it knows, and where it should escalate. Plain language, the way you would brief a new hire.
AI support agents
An AI support agent built into your product.
Drop one script tag onto your site or app and your customers get an agent that answers questions, fills in forms, and captures bug reports, in your brand, around the clock. Not a chatbot that deflects. A support agent that does the work.
How it works
From a script tag to a working support agent.
No SDK to wire up, no model to host. You brief the agent, drop it on the page, and it starts doing real support work inside a sandbox.
One line on your page.
A single script tag loads the agent in your brand. It reads the page it sits on and is ready to help your visitor.
Answers, form-fills, reports.
It answers from your knowledge, fills in forms through signed webhooks, and routes bug reports to your queue. The agent has no network of its own, every action goes through GATE.
Signed and logged.
Every conversation and every tool call is logged and signed. You can see exactly what it said and did on your behalf.
In practice
What it looks like on your product.
Real tickets, answered from your own docs, resolved in the session or escalated with a clean report.
"Does your plan cover more than one site?"
In frontend mode it answers from your own docs and, if the visitor is interested, captures them as a qualified lead, so a pricing question becomes a lead instead of a bounce.
"Switch my plan to annual."
In backend mode it makes the change through a signed webhook, right there in the session, so a logged-in customer gets it done instead of waiting on a ticket.
"This button does nothing when I click it."
When something is genuinely broken, it files a structured report to your queue, the page, the steps, and what failed, so your team gets a real ticket instead of a vague complaint.
"What is your refund policy?"
It answers only from your own help docs and links the article, and never invents a policy you don't have, so every answer is one you would stand behind.
What you get
An agent, not a widget.
Sales, signed-in help, or bug intake
Frontend mode answers anonymous visitors and pre-qualifies leads. Backend mode helps logged-in users and acts through signed webhooks. Report mode captures structured bug reports. Same install, three jobs.
Your colors, voice, and knowledge
The agent runs in your brand skin and answers only from the knowledge you give it, so it never invents a policy or contradicts your docs.
No network, no secrets on the page
The agent runs with no network access. Every action flows through a GATE socket, so the customer's browser never sees a key or a token.
Memory of what it supports
Point it at your help articles and docs and it remembers them, so answers stay consistent as your product changes.
Bug reports that arrive structured
When something is genuinely broken, it files a structured report to your queue instead of apologizing in a chat window.
Support data stays in the EU
Runs on Stockholm-hosted, EU-resident infrastructure. Customer conversations are never used to train models.
How it fits
One agent, the whole support job.
The installable agent is the front line. Behind it, the same fleet runs your customer support queue, answers from your company files, remembers your product through the platform's memory, and gives your team the same brain inside Slack.
For the full picture, read how an AI agent runs your support queue.
FAQ
Straight answers.
Is this just a chatbot?
No. A chatbot hands a customer a help article and hopes they leave. This is an agent that reads the page, answers from your knowledge, fills forms through signed webhooks, and files structured bug reports, all from a sandbox.
Do customers need to log in?
No. Frontend mode helps anonymous visitors and pre-qualifies leads. Backend mode helps logged-in users and can act on their behalf through signed tools.
Where does the conversation data go?
Conversations run on EU-resident infrastructure. The agent itself has no network access, every action goes through GATE, and nothing is used to train models.
Drop a support agent onto your product.
Invite-only while we onboard a small number of businesses. Request an invite and we'll be in touch.