How do I do this specific thing?
The documentation answers it. Nobody finds the right page.
Support headcount grows with users unless something absorbs tier one, and that something has to be accurate or it just adds a step before the human.
A queue that grows linearly with signups.
Does the API let me bulk update records? Docs are ambiguous.
Yes, PATCH on the collection endpoint takes up to 500 in one call. The rate limit is 100 requests a minute on your plan. Here is the page.
cited the API referenceI get a 422 with what looks like a valid payload
That is beyond what I can diagnose from here. I am opening a ticket with your payload, the endpoint and your plan attached so support does not have to ask for it. You should hear back in the morning.
stopped guessing and escalated with the detailNot a category list. These are the messages that arrive, and the reason each one is harder than it looks.
How do I do this specific thing?
The documentation answers it. Nobody finds the right page.
Does it integrate with what we already run?
A pre-sales question that decides the trial, asked constantly.
What is the difference between these plans?
The comparison that stalls an upgrade.
Is my data safe, and where does it live?
Security review, arriving early and from every serious buyer.
Why is this not working?
Sometimes documentation, sometimes a real bug. Telling them apart is the job.
Can I get an invoice or change my plan?
Billing admin. High volume, zero judgement required.
Is this a bug or am I doing it wrong?
The one that has to reach an engineer with the details already attached.
The questions that decide whether a trial converts arrive before anyone has been assigned to the account.
Nobody freeA global product gets support questions around the clock and a small team does not.
Nobody freeThe same handful of questions, hundreds of times, from people who did not read the changelog.
Nobody freeThe account that quietly stopped asking questions is the one about to leave.
Your docs, changelog and help centre become answers to the question as it was asked, with the page cited so the customer can go deeper.
Shows where the answer came fromIt resolves what it can and escalates what it cannot, rather than paraphrasing the same page three times. The second behaviour is what makes customers hate support bots.
Hands over to a personWhen it does escalate, the version, the steps and what the customer already tried are attached, so the first reply is not a request for information.
Opens a ticketIntegrations, limits and security, answered during the trial rather than in a reply two days later when the evaluation has moved on.
Captures the enquiryEvery trade has a line an agent must not cross, and the reason is usually law, insurance or a licence rather than good manners. This is configured before anything else.
A confident wrong answer about a technical fault costs more than no answer. It escalates with the detail attached instead.
Those answers are commitments. It points at what you publish and routes the rest to the people who sign them off.
Money and entitlements stay with you. It gathers what is needed so the action takes seconds.
Every row is a job somebody does today. The bar is counted from the rows, so it says exactly as much as the table does and no more.
Everything listed is live today. See the full list.
We would rather say so than borrow a customer from a different trade. These are live and you can talk to any of them right now, then judge whether the same thing would answer your enquiries.
See the live agents→Your CX Agent, set up to deflect and escalate.
An AI support agent trained on your own content that answers the repeated questions with citations and hands the rest to a person with the full transcript.
See Support Agent→Each of these is something a customer of yours gets, rather than a feature you switch on. Where it comes from is marked, so you can see what arrives with the agent and what needs a channel or your store connected.
submit_ticket Tool call Called when the agent cannot resolve something. It attaches the transcript and the detail already gathered, so nobody has to ask again. Escalates with the detail already attached, so the first reply is not a request for more information. handoff_human Tool call Called when a person is needed, by rule or by judgement. It passes the whole thread across rather than a notification. Passes the whole thread across, so the customer never has to explain themselves twice. retrieval Built in Every answer is drawn from your own content and carries the page it came from, which is what makes grounding checkable rather than claimed. Every answer names the page behind it, so you can check it and so can your customer. lead_capture Tool call Called when intent appears. It records who asked, what they wanted and how ready they are, while the conversation is still open. Records who asked, what they wanted and how ready they are, while the interest is still warm. The one you tried almost certainly paraphrased a help page and would not admit when it was stuck. This cites the page it used, and escalates with the detail attached instead of trying a third rewording.
Good. Size is where retrieval wins, and re-reading on a schedule is the point. Publishing an update updates the answers.
It answers from your documentation and cites it. A question the docs do not cover becomes an escalation, not an invention. Grounding is checkable in every conversation.
That tends to be where it pays first. Integration, limit and security questions during a trial are the ones that decide whether the trial converts.
Yes. It is the same agent wherever you put the widget, with the same knowledge.
Paste your URL and OU Chat trains on your own pages, so the first answers it gives are about your business rather than a template. Seven days without a card, $29/mo after that, and a full refund on any plan within 30 days if it is not right.
Build your agent→