We started GeFi because none of the existing AI marketplaces are structured
to serve a regulated financial-services buyer. They sell you a model. They
don’t tell you where the training data came from, where the inference ran,
or how to prove either to your regulator.
So we built one that does.
What’s live today
- Marketplace — the catalogue is small but real. Five featured models
across NLP, optimisation, credit risk, fraud, and trade finance.
- Audit log — every inference is hash-chained and Merkle-anchored
daily. Spec is on GitHub.
- Federated training — two production federations (Credit Oracle and
Portfolio Optimiser) with 41 combined institutional participants.
- Compliance routing — KYC tiers and per-jurisdiction counsel are live
for US, UK, EU, and UAE.
What’s next
- More models. Macro Nowcast comes out of beta in Q3. Treasury
liquidity model and ESG attribution model are in private alpha.
- MAS and FINMA. Singapore and Switzerland counsel onboarding starts
this quarter.
- Sovereign tenants. OVH SecNumCloud and Infomaniak deployments are in
staging.
- Public trust portal. trust.gefi.io goes live
with the SOC 2 Type I report.
What we’re still arguing about
- Onchain payouts as the default. It’s cleaner for global developer
payouts; it’s a UX cliff for fiat-only buyers. We’re shipping both rails
and watching what people use.
- Open-sourcing the model evaluation harness. We want to. We don’t want
to do it before it’s clean. Probably Q4.
- A native EU Cloud Region beyond Cloudflare. Cloudflare’s EU region is
good. It is not “we run on a SecNumCloud-qualified provider” good. The
Enterprise tier ships with that option; the question is whether we make it
the default for EU paid tenants.
If any of this is interesting, you can request early access (the self-serve sign-up at app.gefi.io opens with our next release)
or book a demo. If it’s very interesting,
we’re hiring.
— The GeFi team
]]>