I started building this after noticing something that kept bugging me: AI agents are getting genuinely good at *doing* things searching, comparing, booking, researching. But the second one of them needs to actually pay for something, it just stops. Someone has to step in and click approve.
That felt backwards. If an agent is trusted enough to do the task, why can't it be trusted to spend $2 on an API call it needs to finish that task?
So I started digging into how payments actually work — bank transfers, cards, UPI, all of it — and realized every single one of these systems was built with one hidden assumption: a human is always the one hitting confirm. Nothing was built for a world where that's not true anymore.
Orite is my attempt at fixing that. Not by replacing how payments move, but by building the missing layer in front of it the thing that decides *whether* a payment should happen at all, based on limits you actually set, before any money moves.
The approach changed a lot while building this. I originally thought this was just "give the agent a payment API key." It's not. It's closer to giving an employee a company card limits, categories, an audit trail, a way to say no. That reframing changed almost everything about how I built it.
This is still early a first version, built solo, and I'm sure there are rough edges. Would genuinely love to hear what you think, what's missing, and whether this is a problem you've run into building your own agents.
Thanks for checking it out 🙏
PH 用户
You asked what is missing. The spend that sits inside the limit and is still wrong. Right amount, confident agent, wrong thing. A budget does nothing there. Can autonomy options be set till the time trust gets built with user?
Also can I reverse after the fact or raise dispute (worst case scenario)? And does the log/trace tell me which decision led to the charge, or only what the charge was? That second one is what lets me go fix the agent right, which is gonna be criitcal.
I started building this after noticing something that kept bugging me: AI agents are getting genuinely good at *doing* things searching, comparing, booking, researching. But the second one of them needs to actually pay for something, it just stops. Someone has to step in and click approve.
That felt backwards. If an agent is trusted enough to do the task, why can't it be trusted to spend $2 on an API call it needs to finish that task?
So I started digging into how payments actually work — bank transfers, cards, UPI, all of it — and realized every single one of these systems was built with one hidden assumption: a human is always the one hitting confirm. Nothing was built for a world where that's not true anymore.
Orite is my attempt at fixing that. Not by replacing how payments move, but by building the missing layer in front of it the thing that decides *whether* a payment should happen at all, based on limits you actually set, before any money moves.
The approach changed a lot while building this. I originally thought this was just "give the agent a payment API key." It's not. It's closer to giving an employee a company card limits, categories, an audit trail, a way to say no. That reframing changed almost everything about how I built it.
This is still early a first version, built solo, and I'm sure there are rough edges. Would genuinely love to hear what you think, what's missing, and whether this is a problem you've run into building your own agents.
Thanks for checking it out 🙏
Also can I reverse after the fact or raise dispute (worst case scenario)? And does the log/trace tell me which decision led to the charge, or only what the charge was? That second one is what lets me go fix the agent right, which is gonna be criitcal.