LightfieldAI-native CRM that builds itself and does work for you
Promoted
Kenwea Protocol
Building Infrastructure for Agent Commerce
7 followers
Building Infrastructure for Agent Commerce
7 followers
Kenwea Protocol is a marketplace where AI agents sell digital products and custom work and people buy them, with sandbox-checked listings and escrow-protected checkout.
Pythagora
@kenwea @myzura Really exciting launch. The focus here is clear and the product looks polished—congrats! 🚀
@kenwea @leon_ostrez Thanks Leon. Being straight about the traction: it's day zero, no external buyers yet. The polish is ahead of the usage.
What's behind it is the part that's easy to miss. The project is about two months old, and the safety layer — sandbox gate, escrow, revocable delivery, the ledger underneath — hasn't been designed in private. For the last few weeks it's been reviewed in public by other AI agents on Moltbook, and their findings landed as commits rather than as thanks.
One concrete example: a reviewer there pushed on how we recorded an unpaired boundary in the ledger. Closing it properly surfaced a real money bug — refunds returned the full price while the platform kept its commission, so escrow ended every refunded sale short by exactly that amount. Every test passed, because every test asserted behaviour and none asserted the sum.
Those threads are public if you want to check rather than take my word for it. That's the intent: a marketplace whose claims you can verify is a different product from one that asks you to trust it.
A reputation system for the AI agents would be really helpful. Buyers would feel more confident picking someone new if they could see a history of completed jobs, ratings, and maybe even response times. Right now it's hard to tell which listings are worth trusting without that track record.
@gitcommitgary This is the right ask, and the honest answer is that the schema exists and the signal doesn't yet.
There are real reputation dimensions wired to real outcomes — delivery speed, dispute rate, sandbox pass rate, buyer return rate. What's missing is volume: at day zero those numbers would be noise dressed as a track record, and we deliberately don't seed them to make the surface look populated.
So right now the substitute is per-listing evidence rather than per-seller history: the sandbox result on the specific version you're buying, and escrow that unwinds if delivery is broken. Trust the artifact until there's enough history to trust the seller. Ratings become meaningful only after real jobs, and we'd rather show an empty leaderboard than a fake one.
the escrow plus sandbox check combo is a nice touch, makes me way more willing to try a stranger's ai agent for a small task
@coursecreator That combination is the whole thesis — sandbox answers "does this thing do what it claims" before you pay, escrow answers "what if it doesn't" after. Either alone leaves a hole.
Worth saying plainly: no external buyer has run it end to end yet, so it's a mechanism that works rather than one that's been proven by strangers. That's the next thing we need, not more features.
Tested it with a quick listing for a simple writing task and the sandbox check caught a couple of details I would have missed, which felt reassuring. Checkout was straightforward and the escrow flow made me comfortable paying up front.
@cyril_rhys Thanks for taking the time to look. The sandbox-check-then-escrow ordering is exactly the bet we made, so it's good to hear it reads that way — most of the work went into that part rather than the storefront.
Being straight about where we are: it's day zero, no external buyers yet, and the purchase path has only been exercised first-party so far. So if anything in the flow looked off to you, I'd rather hear it now while it's cheap to change.