Pricing and billing
How much does this cost, and what's included?
$99 a month, no setup fee. That covers a full crawl of your published help content, an agent trained on it with real citations back to your own articles, and ongoing re-checking so it stays accurate as your docs change. It's one price, not a tiered ladder. Everything above is included, not an upsell.
Is there a contract, and when am I actually charged?
No contract. It's month to month, cancel any time, no multi-month minimum, no early-termination fee. The first month is due at signup, before onboarding starts, then billed monthly in advance after that. Cancelling stops future charges, it doesn't refund the month already paid. The trade-off is honest: because there's no lock-in, we have to keep being worth the $99 every month, not just at signup.
If I cancel, what happens to my widget and my data?
The widget stops answering and comes down (or we tell you how to remove it, if you added it yourself) as soon as you say so. If you want a copy of what we built, your indexed content and a usage summary, we'll give it to you. We don't keep a record of what your visitors asked, so there's nothing beyond that to hand over. We keep our own copy of the underlying software, since that's the same system every customer runs on, but nothing about your organization stays exclusive to us once you leave.
What happens if we get way more traffic than expected, do I get a surprise bill?
No. There's no per-question or per-resolution overage charge on this plan, which is a real difference from tools priced per resolution. Each subscription includes a volume allowance sized comfortably above normal usage for a business your size. If your real traffic looks like it's approaching that, we'll reach out to raise it or talk about pricing before it becomes a problem, not silently meter you and send a bigger invoice.
Do you work with nonprofits? Is there nonprofit pricing?
Yes. We've built and donated agents for a handful of nonprofits already, at no cost. There's no separate published nonprofit rate. Email us, tell us about your organization, and we'll figure out what's fair on a case-by-case basis rather than promise a fixed discount tier that may not fit your situation.
How it actually works
Do I need a developer to install this?
Usually not much more than you'd need to add a Google Analytics tag. It's one script tag pasted once into your site, no build step, no app to install, no account for your visitors to create. Someone still needs the ability to edit your site's template or use a tag manager, the same as any other embedded widget, but there's no ongoing developer involvement after that.
Does it need to integrate with our CRM, helpdesk, or ticketing system, or run on our servers?
No to both. It doesn't touch your ticketing queue, CRM, or existing live-chat tool, and it doesn't require one to exist. It's a standalone answer box trained on your public docs. If you already run a helpdesk widget, this can sit alongside it; it isn't asking you to replace anything or migrate tickets anywhere. It also runs entirely on our infrastructure, not yours. Your side is the one script tag, and everything else (crawling, indexing, the agent itself) lives on our end. If a customer's compliance requirements genuinely require it to run on their own infrastructure instead, that's a different, higher-touch arrangement we can talk about, but it isn't the default and isn't what most customers need.
What happens when our docs change, does the agent just go stale?
It only knows what's already published. It isn't watching your site live, word for word, in real time. We check your docs for changes on a regular schedule as part of the monthly service and update the agent automatically when something's different. That's included, not a paid add-on you have to ask for.
How do you make sure an update doesn't quietly make it worse?
Every update has to pass the same accuracy check the very first version had to pass before it's allowed to go live. If an update doesn't pass, it doesn't go out automatically. A person looks at it first, instead of an untested change reaching your visitors. Nothing changes on your live widget just because we ran something; it changes because the update earned it.
Trust and quality
How do you keep it from making things up?
Two layers. First, it only answers from your own published material, and says plainly when it doesn't know rather than guessing. Second, we don't just take its word for it. Every link it gives is automatically double-checked afterward against the real article it's supposed to point to, and anything that doesn't match gets fixed or removed before your visitor ever sees it. A broken or made-up link simply can't reach someone using the agent.
What happens when it genuinely doesn't know the answer?
It says so directly, in plain language, and points to a real person, your support inbox or whatever escalation path you give us, instead of sending someone back to go search your help site themselves (they're already there, asking). It won't tack on a source link to an answer it didn't actually give. That redirect is a feature working correctly, not the agent failing. It's trained to decline rather than guess, so if you see it hand off to an email address while testing it yourself, that's the safety behavior doing exactly its job.
Do you see the questions our customers ask?
No. We don't log or store the questions your visitors ask, or the answers the agent gives back, nothing to review, nothing to export, nothing sitting in a database. The job here is simple: answer well enough, right on the spot, that fewer of those questions turn into an email or a support ticket. That's the whole value, not an ongoing feed of your customers' conversations.
Is our data, or our customers' data, sold or shared with anyone?
No, never, for any reason, and there's nothing to sell in the first place. What it's trained on is just your own already-public help content, so there's nothing sensitive there to begin with, and we don't log or retain your visitors' questions or the agent's answers at all. We don't use your content to train any other general-purpose model either.
What if it gives a wrong answer and someone acts on it?
We test for this before anything goes live and before any update goes out, but no language model is perfect, so we don't claim otherwise. Every answer includes a clear way to reach a real person, and it's built to decline rather than guess when the material doesn't support a confident answer. For anything safety-critical, financial, medical, or otherwise high-stakes, treat it the way you'd treat any AI tool: a first answer to verify, not a final one, and say so to your own visitors if that matters for your situation.
Security, technology, and reliability
What AI model or technology powers this?
We build on established commercial AI providers rather than a model we trained ourselves, and we pick the specific one based on cost and accuracy for this kind of grounded question-answering, that choice can change over time as better options come along. What stays constant is the guardrail layer around it: it only answers from your own published docs, links every claim back to a real source, and every update has to pass an accuracy check before it goes live. If your organization needs a specific provider or region for its own compliance reasons, that's part of the same higher-touch, run-on-your-infrastructure conversation mentioned above.
Are you SOC 2 or ISO certified?
Not yet. We're a small, early-stage service, and formal certification isn't something we can honestly claim today. If a certification is a hard requirement for your organization, tell us up front and we'll say plainly whether we're a fit right now rather than string you along.
Are you HIPAA compliant, or can you handle regulated or sensitive data?
Not by default, but for most people asking, that matters less than it sounds like it should. The agent only answers from your own already-public help documentation, and it isn't a place where patient or health data belongs in the first place, so HIPAA usually just isn't the right lens for how this gets used. If your organization genuinely needs it anyway (some procurement processes require it as a checkbox regardless of the actual data involved), that's a higher-touch, run-on-your-own-infrastructure conversation; email us and we'll tell you honestly whether it's something we can support for your case.
Where is data hosted?
On cloud infrastructure in the United States. Data residency is usually a concern for tools handling personal or sensitive visitor information, this one only reads your own already-public help content and doesn't collect or store anything about your visitors, so it typically isn't the same kind of question here. We don't offer region-specific hosting (EU-only, etc.), so if that's a hard requirement regardless, this likely isn't the right fit today.
Do you offer an uptime guarantee or SLA?
Not a formal one with credits. We're a small, early-stage service and would rather say that plainly than promise something we can't back up operationally yet. In practice we watch for problems and fix them quickly, but if a contractual SLA is a hard requirement, tell us and we'll be straight with you about whether we're a fit today.
Does it show your branding, or ours?
Yours. Once it's live on your site it carries your name and your colors, the "AI Support Agent" branding you see on our own demo is just us running our own product on our own homepage, the same way any of our customers run it under theirs.
Does it work in languages other than English?
It can generally attempt an answer in whatever language a visitor asks in, but we've only rigorously tested it against English-language documentation. If answering in other languages matters for your site, ask us for a real test on your own docs before committing, same as the free example we build for every prospect.
Can one subscription cover multiple sites or brands?
Not by default. Each subscription covers one site's documentation and one embedded widget. If you run multiple brands or properties and want to talk about covering more than one, email us and we'll work out something fair rather than make you buy duplicate subscriptions.
How this compares to other options
How is this different from Intercom Fin, Zendesk AI, or a tool like Tidio?
Those are full support platforms, ticketing, live chat, a shared inbox, staff workflows, with an AI answer feature built on top. To get the AI piece, you generally have to adopt the whole platform underneath it, migrate your tickets, and get your team trained on a new inbox. This isn't a platform at all. It's one script tag that answers questions from your existing docs; there's nothing to migrate and nothing new for your team to learn. On raw price, this is also simply cheaper than most of those specific AI features on their own. Intercom Fin runs roughly $297 to $495 a month at typical volume, and Certainly is around $449 a month, but the bigger difference is that you're not buying a platform migration to get there.
Why wouldn't we just build this ourselves with a DIY chatbot tool?
You can, and some of those tools are fine starting points. But the cheap entry tiers (commonly in the $30 to $40/month range) usually come with the vendor's own branding on the widget, capped monthly volume, and manual updates only. A real like-for-like setup, once you add branding removal and automatic content updates, tends to land closer to a $99/month median and climb from there. And critically, someone still has to keep it accurate as your docs change, that ongoing maintenance is the actual job, and it's the part a self-serve tool doesn't do for you. That maintenance, with an accuracy gate in front of every update, is what the $99/month here is actually paying for.
Why isn't this a bigger platform with more features?
Because that's a deliberate choice, not a limitation we're working around. Getting the AI piece usually means adopting a whole new system underneath it first. This asks the opposite of you: you don't replace anything you already have, you add one thing that answers a question your docs already contain. Simple is the product, not a stepping stone to a bigger one.
Before you commit
We don't have a lot of documentation. Is this even worth it for us?
Sometimes, honestly, no, and we'll tell you that up front rather than take your money and find out later. The free example we build on your own docs before you decide anything is exactly how we check: if your published material is too thin to ground good answers, that's useful information either way. It usually means the real opportunity is thickening your docs first, not buying an agent yet.
What if our docs are outdated or have mistakes in them?
The agent can only answer from what's actually published, so if something in your docs is wrong or stale, the agent will reflect that, it doesn't independently fact-check your content. That's fine for what this is actually meant to do: give a good-enough answer, right away, so fewer of your customers' questions turn into an email or a support ticket, not continuously audit or improve your documentation. If you notice something's wrong, fix the doc yourself, and the next scheduled update picks up the correction automatically.
Can we see it working on our own docs before paying anything?
Yes, that's the intended starting point, not an afterthought. We build a real, working example using your actual published help articles first, so you can ask it your own hardest questions and judge the answers yourself. No commitment and no sales call required just to see it. If you want to send us your five toughest real support tickets before deciding anything, we'll run those too and show you the answers unedited, including the ones it gets wrong.