An inference cloud that can show its work

For EU mid-size company IT/data-protection leads dealing with AI Act traceability

Most "AI Act ready" vendor claims are marketing gloss on top of the same opaque hosted-inference stack everyone already uses. We'd rather show the actual mechanism than the slide.

The traceability problem, concretely

If your organization runs inference through a third-party API, you typically get: a response, a bill, and not much else. Which physical machine ran it, what else that machine was doing at the time, whether the same weights that were audited are the ones actually serving your traffic — none of that is usually visible to you, let alone provable to a regulator or an internal auditor.

What cloud-murakumo does differently

cloud-murakumo keeps every inference run as an append-only, actor-signed ledger entry: which node served it, which model weights (content-addressed, not just a version string), and how many GPU-seconds/tokens it consumed. This isn't a compliance checkbox bolted on after the fact — it's the same ledger the billing and autoscaling math already run on (you can watch it compute live in the playground; the scheduler you see in the browser is the identical code the CLI and the production fleet use).

We're not claiming this alone makes you AI Act compliant — that's a legal determination for your own counsel, not an engineering one. What we're claiming is narrower and verifiable: the underlying traceability primitive (node → weights → run, byte-checkable) exists and is the same one production traffic uses, not a separate "audit mode."

Two ways to use it

1. Keep inference inside your own overlay. Tokens don't leave your organization's network boundary by default (default-deny policy) — useful if data residency/sovereignty is part of your compliance posture, not just traceability.
2. Own the hardware. #store/eu sells GPU/PC node ownership shares — buy the node, it joins the fleet, you get a compute-revenue lease on it. Several EU member states allow accelerated depreciation on IT hardware (Germany's near-immediate write-off for computer hardware/software is one concrete example) — that's a tax question for your own advisor, not something we're asserting compliance on.

Payment is Stripe, card only for now (no installment financing at this time). Crypto payment is not live — it's a disabled placeholder on the storefront, not a promised timeline.

Who this is actually for

Mid-size companies that already have to answer "can you show us how this model call was served" to a customer, auditor, or internal risk team — and currently can't, because their inference vendor's answer is "trust our SOC 2 report." We're offering something narrower and more mechanical: the actual run ledger, not a compliance narrative.