Meerlume

Me and my co-founder just launched Meerlume - you describe your business in plain language and get a booking bot on WhatsApp or Telegram.

The interesting part isn't the generation. Everyone does that now. It's where we put the model.

Every competitor in this space runs an LLM on every incoming message. That means their cost scales with your customers' chattiness, which is why they all charge per message or per "AI credit." It also means the bot can say things the owner never approved.

We compile instead. The model reads your description once and produces a flow graph. Editing runs through the model too. But the thing that talks to your customers is a static graph with no inference in it.

What that bought us:

- Marginal cost per conversation is basically zero, so the free tier can be real instead of a 25-contact tease

- The bot cannot invent a price or a policy, because there's nothing generative in the runtime path

What it cost us:

- The bot genuinely cannot handle "do you also do X?" if the owner didn't put X in the flow

The thing I'd tell anyone considering the same trade: the deterministic path is harder up front and much cheaper forever. Whether that's right depends entirely on whether your users need a conversation or a transaction. Ours need a transaction.

Happy to answer anything about the architecture or the numbers.