Built for the people running more sites than they have hours.
A small team, an unfashionable customer, and one architectural decision that everything else follows from.
Hepo started because the tools for this customer were all built for someone else. Enterprise help desks assume a support department. Free chat widgets assume one website and no budget. Neither assumes a person in one timezone selling into another, running six sites from a laptop, answering in a language they did not grow up speaking.
That person exists in large numbers and buys almost entirely on recommendation from other people like them. It is a good customer to have and a hard one to fool, which suits us.
The decision everything hangs on
Chat, session recording and website analytics share one visitor id. Not joined afterwards. The same row, from the first schema we wrote.
It sounds like a detail and it is the whole product. It is why the recording is already attached when a message arrives, why analytics can tell you whether people who talked to you bought more, and why nobody can catch up by shipping an integration.
What we are not building
Not an enterprise help desk. That market is won with procurement processes and a sales team, and the product comes third. Not a marketing automation suite. Not a platform, yet: the architecture supports installable applications and we will say so when there is one worth installing, not before.
Four things we decided early
Charge for the expensive thing
Seats cost us a table row. Model tokens cost us money that shows up on a graph. Billing the second and not the first is the only version that does not push customers into sharing a login.
One record, not an integration
Chat, recording and analytics were written against the same visitor id from the first commit. Everything interesting this product does falls out of that decision, and none of it can be added later by writing a connector.
A tool you can take with you
Exports on every plan, in formats that work without us. A binary you can run yourself. If leaving is hard, staying was never really a choice.
Say the unfinished parts out loud
Our security page lists what we have not certified. Our email page says which providers are still blocked on a review. People find out anyway, and finding out during evaluation is cheaper than finding out after.
The closed beta is still open
We are running a small number of real sites rather than a public launch, because a support system only shows its problems under real traffic. Dropped connections, message backfill after an outage, corporate proxies cutting sockets, recordings degrading on a site with an awkward CDN. None of that appears in a test environment.
If you run a site with real daily traffic and you are willing to tell us when something is broken, we would rather have you than another blog post.
Put it on one site tonight.
One script tag, no card, unlimited seats. If it does not earn its place by the weekend, delete the line.
Free plan stays free. The AI is the only thing that ever costs you anything.