Your AI support agent, inside your product.
Moss sees the screen your customer is on, walks them through it, files what’s left — and proposes the fix so the question stops coming.
Book a demoMoss beat the video tutorial on every metric — upload rate, activation, and retention.
“The test is a success. We built the interface to get people to upload, and they really do upload more.”
“For academic and especially enterprise users, the difference is enormous.”
“Users who had a good onboarding experience came back and uploaded again at a higher rate.”
01Customer Experience
Less time for customers to get unstuck.
The answer arrives on the screen where the question happened, so nothing waits in a queue.
Read the chatbot solution02Product Growth
Build onboarding faster, for every account.
Authored from your docs and repaired when the UI changes, so there is nothing to build by hand.
Read the walkthrough solution03Customer Support
Save the cost of a growing support team.
Every session is scored against criteria you wrote, and the failures arrive already grouped by cause.
Read the analytics solutionWhat is in it
Five capabilities, one agent.
Three that meet your user on the screen, two that bring your team the next change.
On-screen customer support
Answers with sources on the screen where the question happened, then points at the next click.
AI walkthroughs
Authored from your docs, or generated live when nothing covers the flow.
AI agent procedures
Plain-English procedures become agents that check your systems, act, and hand off.
AI session monitoring
Checked against criteria you wrote, with the reasoning and source sessions attached.
Copilot
It finds the pattern, shows the evidence, and helps you update the system.
Autopilot
Find what needs to change without an investigation.
Every session above feeds one loop. Autopilot turns the patterns into the next change your team should make—with the evidence to act confidently.
Moss ships improved versions of itself: the same agent moves from version one to version two to version three, each upgrade made by Moss on its own, every day.
How Moss compares to the way teams solve this today.
Every approach below is useful. Moss is the only one that lives inside your product.
| What it can do | Moss | Chatbots (Fin, Decagon) | Static tour apps |
|---|---|---|---|
| For your users | |||
| Sees the screen the user is on | Yes· live DOM & screen | No | No |
| Takes actions for the user | Yes· with your approval | No | No |
| Goes live in minutes | Yes· one SDK line | Partial | No |
| Keeps working when the UI changes | Yes· reads at runtime | Partial | No |
| For your team | |||
| Scores every session against your criteria | Yes· with the reason | Partial | Partial |
| Answers questions about your sessions | Yes· ask in plain English | No | No |
| Improves itself from every session | Yes· a change a day | Partial | No |
Frequently asked questions
How does Moss work?
Moss is an AI support agent that lives inside your web product. It sees the screen your user is on, answers questions with sources from your documentation, walks the user through the steps on the live UI, and takes permitted actions like filing a ticket. Then it reviews every session, finds what keeps failing, and proposes fixes your team approves — so it improves every day.
How easy is the setup?
One line of the SDK embeds Moss in your product; then you connect the documentation it should answer from — your help center, docs, or uploaded files. There are no flows to build and no UI elements to map, because Moss reads the screen at runtime.
Can we test Moss before our users see it?
Yes, at two levels. Before adding any code, share a demo environment and we will run Moss on it and send you a recording of it working your real flows. After install, every procedure runs in a sandbox first, so you can dry-run and edit it before publishing.
How accurate is Moss?
Accuracy is enforced, not assumed. Every answer must cite a source from your documentation, and every action follows a procedure your team wrote, tested in a sandbox before it goes live. Then Moss checks its own work: every session is reviewed against criteria you define, and the ones that failed are flagged with the reasoning attached — so a wrong answer becomes a fix, not a pattern.
Can Moss take actions for the user, not just highlight?
Yes. When the answer is “here is where to click,” Moss highlights the control on the live screen. When the user wants the task done, Moss can fill the form or file the ticket itself, within the permissions you set. You can mark any element off-limits — a “Pay now” button, for example — and Moss will guide there but never click it.
Does Moss hand off to a human or connect to our helpdesk?
Yes. You define in plain English when Moss should answer, when it should act, and when it should stop and hand off to your team. Today it files Zendesk tickets with the conversation summary and context attached, and reads knowledge through Intercom and Document360. Ask us about your helpdesk — connectors are expanding.
Can Moss publish changes on its own?
No change publishes silently. Moss prepares each proposed change as a diff with its evidence attached, and someone your team designates approves, edits, or dismisses it. Publishing permissions are yours to set, and every published change is versioned, so you can always restore a previous state.
Where can Moss be deployed?
Three ways: in Moss's cloud, in your own cloud, or fully on-premises, including air-gapped environments. On-premises runs in your infrastructure, so data provided to Moss stays inside your perimeter.
How does Moss use our data?
Your data is not used to train any model, ours or a provider’s. Redaction is on by default, and screen data can be purged on processing with an audit log to verify it. Our declared model sub-processors are OpenAI, Anthropic, Google AI, OpenRouter, and Baseten. SOC 2 and ISO 27001 are both in progress, and a penetration test was completed in May 2026.
See Moss inside your product.
We will set it up on a sandbox of your actual product during the call.