Runs on Google Calendar, MongoDB Atlas, OpenAI, Vercel, AWS and Resend
- Vercel
- MongoDB
- OpenAI
- AWS
- Resend
What you can point it at
You have already written the answers
What your assistant does for you
Answers at 2am
Out of your own documents, with the source shown under every reply. When your files don't cover the question, it says so and takes a message.
Books on your real calendar
It reads your Google Calendar free/busy, holds the slot, and emails the confirmation to both of you. Not a link — the booking.
Captures the lead
Name, need and contact details, asked one question at a time and written to a record you own. Nobody has to remember to follow up on Monday.
What that looks like in practice depends on what you do — there are pages for clinics, consultants, studios, trades and creators, each naming the paperwork that line of work already has.
They ask. It answers. It books.
Your documents on one side, a customer's question on the other, and a slot held on your real calendar before they close the tab.
Read, chunked and indexed automatically — no folders, no tags.
The slot is held the moment it's offered — nobody else can take it.
Feed it what you already wrote.
Drop the files in, forward them to your private capture address, or pick them out of Drive. Uploads land straight away and keep processing in the background, so you can close the tab on a fifty-page PDF and get on with your afternoon.
- PDFs, images, DOCX, screenshots
- Forward an email straight into your knowledge base
- Import from Drive without granting your whole Drive
It reads the pages, not the titles.
Every document is extracted and split into passages, so a question about clause three of your cancellation terms can actually reach clause three. What you end up with is an assistant that quotes your policy rather than paraphrasing the gist of it.
- Text, tables and scanned pages
- Dates, renewals and amounts detected
- Filed without folders or tags
Ask it what a customer would.
Two searches run on every question — one for meaning, one for the exact words — and the results are fused. So “do you take Delta?” and “insurance accepted” land on the same page of your policy, and the phrase a customer happens to reach for stops deciding whether they get an answer.
- What's your refund window?
- Do you work with clinics?
- How much for a 90-minute session?
It books the meeting.
When someone is ready, the assistant reads your real Google Calendar free/busy and holds the slot in the same moment it offers it. Two people cannot end up in the same half hour — the second booking is refused outright, rather than found by you the next morning. Google Calendar only, today.
You can check its work.
Open any conversation and you can see the passages an answer was built from, what each one scored, and the floor they had to clear. Save the questions your business cannot afford to get wrong, and we re-check them whenever your knowledge changes.
- A receipt kept for every public answer
- Gaps clustered by topic, not one long log
- Checks watch retrieval, not correctness
Your knowledge stays yours.
Every search is filtered by your account id inside the database query itself. That filtering happens before any model is involved, which makes isolation a property of the index rather than an instruction another prompt could talk its way around.
The same vault runs your life.
This began as a memory for your own documents and it never stopped being one. Insurance renewals, licences, registrations, filings — the assistant answering your customers is the same one that emails you before something expires quietly in a drawer. By email, which is the only channel we have built.
We'd rather you trust us than be impressed by us.
Trust is usually sold with other people's logos. We don't have any yet, so here are four things you can check in the product instead — and, underneath them, the list of what it still doesn't do.
Never trains a shared model
Your documents and your customers' conversations are used to produce your answers. That is the entire list of what they're used for.
Leave whenever you want
Export everything as JSON in one click. Delete your account and it cascades — documents, conversations, assistants, leads.
You control who can embed it
The widget only runs on domains you allow. Public conversations are gated by a challenge and a short-lived token bound to that assistant and that visitor.
Every public answer keeps a receipt
The passages it used, the score each one got, and the floor they were judged against — kept per answer, visible in the transcript, and included when you export.
What we don't do yet+
- Reminders go out by email. Not SMS, not WhatsApp, not push.
- No team or shared accounts. One login, one knowledge base. You can't invite a colleague yet.
- Voice is web voice. Your assistant can speak and listen in a browser. It does not answer a phone number.
- Booking is Google Calendar only. No Outlook, no Cal.com, no Calendly import.
- Card payments are live in India today. International card processing is pending gateway approval.
- We're new and small. If that's disqualifying, we'd rather you know now than find out in month three.
The full version lives on our security page, and we write out where the alternatives beat us in the comparisons. Ask us where the edges are and we'll tell you — get in touch.
It does more than answer now.
The newest things it can do for you, all live today.
Compare Pro and PowerConnects to your own systemsPower
Sync your catalogue or records in from your API, and let it run live, read-only lookups — an order's status, a stock level — while it answers. 4 assistants on Power.
Meeting types, not just a slot
Separate lengths, hours and buffers per kind of meeting. Guests reschedule or cancel from their own link; you get a prep brief before it starts.
Verified visitors
Optionally confirm a visitor's email with a one-time code. Their conversation follows them across devices, and you see one verified lead per person.
Remembers who came back
A returning visitor doesn't start from zero — built only from what they told it, expiring with your transcripts, ended by a “Not you?” link.
Reads your website
Give it your site's address and it crawls the pages into its knowledge, so what's already published doesn't need uploading twice.
A report every Monday
What your assistants were asked, what they couldn't answer, and which gap is worth filling first — in your inbox, not a tab you forget to open.
The bit you're actually wondering
Will it make things up to my customers?
It answers only from the documents you gave it, shows the source under each reply, and says it doesn't know rather than improvising when nothing relevant comes back. It's still a language model, so it can be wrong — which is exactly why every answer carries its citation and why you get a list of every question it fumbled, grouped by topic.
How long until it's live, and how good is it on day one?
Upload what you already have and publish. It's exactly as good as what you feed it: a five-page policy PDF makes a five-page-good assistant. The gap report then tells you what to add next, and answering a gap turns your reply into knowledge on the spot — so week two is better than week one without you planning it.
Is my client data safe? Does it train an AI?
No shared-model training, ever. Isolation is enforced in the database query — every search is filtered by your account id at the index, not by an instruction in a prompt. Export everything as JSON or delete the account and it cascades. The honest part: your content is processed by a third-party model provider under a no-training arrangement. We don't run our own model, and we're not going to pretend otherwise.
Can I stop it promising things I can't deliver?
Yes, and this is the setting people underestimate. You set the tone, the length, the formality and the language, and you give it an explicit list of things it must never promise — discounts, turnaround times, availability you haven't agreed to. Those controls map onto how it actually replies rather than being a suggestion in a prompt somewhere. When a visitor pushes for something on that list, it declines and offers to put them through to you.
How do I know it's still working in three months?
Because you'll be told when it isn't. Save the questions your business can't afford to get wrong, and we re-run them whenever your knowledge changes — then email you if the document that used to answer one has stopped reaching it. Be clear on what that proves: the right source still reaches the answer, not that the answer is worded well. Grading wording would take a second model on every run, which is the cost the rest of the system exists to avoid.
Who else uses this?
Nobody you've heard of. We're new, we have no logos to show you, and we're not going to invent any. What we'll do instead is give you seven days of full Pro, no card: build the whole thing, publish it, and watch a real customer talk to it before you pay anything.