# cloud-murakumo — Terms of Service

> **DRAFT — This is not legal advice. Review by qualified counsel is required before publication.**

Last updated: 2026-08-30

Governing law: Delaware law, subject to mandatory consumer-law rights.

Site / service operator: **Kotoba Labs Inc** (the "Operator", "we", "us").
Sales contact: **Ryo Awai**. Contact: support@murakumo.cloud.
No other entity is the public operator of murakumo.cloud. This repository
does not contain an Inc filing for Kotoba Labs Inc; 法人情報, address, phone
and 代表者 are disclosed on request and are not invented here. Names of the
form `ai-gftd-*` below are product/repository identifiers, not the public
operator.
Service: **cloud-murakumo** — a distributed GPU cloud and LLM/generative
inference infrastructure service (the "Service"), including the public site at
murakumo.cloud.

---

## 1. Acceptance

By accessing the Service, invoking its inference or generation functions
(directly or through an integrating product such as `ai-gftd-router` /
`ai-gftd-apex`), or using the murakumo.cloud site, you agree to these Terms of
Service (these "Terms") and the [Privacy Policy](./privacy.md). If you use the
Service on behalf of an organization, you represent that you are authorized to
bind it. If you do not agree, do not use the Service.

## 2. Service description

cloud-murakumo is a serverless GPU execution and inference platform that hosts
model weights and runs distributed inference and generative jobs. It provides,
subject to configuration:

- LLM serving (e.g. OpenAI-style endpoints served via vLLM);
- generative jobs producing artifacts — image, video, 3D, music, sound effects,
  and voice — through engine runtimes (e.g. ComfyUI, kami-render, TRELLIS, audio,
  TTS);
- model-weight hosting and caching (e.g. via a Hugging Face cache backed by
  content-addressed / object-storage volumes);
- autoscaling, scale-to-zero, scheduled functions, and auction-based placement
  across a GPU fleet; and
- a run ledger recording each inference/generation run.

The Service also permits explicitly consenting smartphone users to contribute
foreground WebGPU/Wasm computation. Participation can consume battery, network
data and device resources, may stop when the operating system suspends the page,
and may be stopped by the user at any time.

MCC is a transferable utility token whose protocol use is purchasing Murakumo
inference. MCC provides no equity, revenue share, ownership, governance,
treasury claim, guaranteed market value, fiat redemption or operator buyback.

The Service is primarily **backend infrastructure** consumed by other products.
Where you reach it through an integrating product, that product's own terms also
govern your use. The Service is offered on an evolving basis; models, engines,
regions, and interfaces may change. The public murakumo.cloud site is a
static/pure-compute surface and does not itself provision GPUs.

## 3. Accounts and identity

- Inference and generation jobs carry an **actor / account identity** supplied
  by the caller; jobs and their run-ledger records are attributed to that
  identity. [CONFIRM: how actor identity is authenticated at the Service boundary
  — e.g. delegated from the calling product, API key, or a kotoba/DID-scoped
  credential — as the repository does not fix an end-user auth scheme in this
  component.]
- When jobs are dispatched through the distributed queue backed by a kotoba
  graph (e.g. `kotobase.net`), the authentication and tenancy model of that
  graph service also applies.
- You are responsible for safeguarding any credentials used to submit jobs and
  for all activity under your identity.

## 4. Acceptable use

### 4.1 The Service is compute; you are responsible for what you run on it

murakumo sells **GPU compute and inference capacity**. It is not a publisher, a
content library, or a distribution platform. We do not select, curate, or
promote what customers generate.

As between you and us, **you are solely responsible for your Inputs and Outputs**
— their legality, the rights you hold in them, the consent of any person
depicted, and your own compliance obligations (including age verification,
record-keeping, labelling, and tax or licensing rules that apply to your
business). You will not rely on us to make those determinations for you.

### 4.2 Absolutely prohibited — no exceptions, immediate termination

The following are prohibited regardless of jurisdiction, purpose, artistic
claim, or whether the depiction is real or synthetic. There is no permitted use,
no research exemption, and no appeal:

- **Child sexual abuse material (CSAM)**, and any content that sexualizes a
  minor, including synthetic or AI-generated depictions and content produced by
  age-regressing an adult subject. We report discovered violations to the
  **National Center for Missing & Exploited Children (NCMEC)** and to competent
  authorities, and we preserve the associated records.
- **Non-consensual intimate imagery (NCII)**, including sexual depictions of an
  identifiable real person created or distributed without that person's consent.

Accounts are terminated on the first substantiated violation of this section
without notice and without refund.

### 4.3 Prohibited

You must not use the Service to generate, process, store, or distribute content
that is **unlawful, infringing, harassing, abusive, defamatory, discriminatory,
or obscene** (as "obscene" is defined by the law applicable to you), or that
incites violence against any person or group. You must not use the Service to:

- generate content you lack the rights to, or that violates a model or engine
  license (including upstream Hugging Face / model licenses);
- impersonate a real person or entity, or create synthetic depictions of an
  identifiable person, without that person's consent — including for fraud,
  political deception, or reputational harm;
- attempt to exfiltrate model weights, bypass quotas or placement controls, or
  disrupt the GPU fleet, scheduler, or run ledger;
- exceed configured GPU-seconds quotas in a manner designed to evade the
  fail-closed accounting and financial-approval gates;
- submit secrets or sensitive personal data into prompts, references, or job
  metadata contrary to the documented data-handling boundary; or
- send unsolicited bulk messages, conduct denial-of-service activity, or
  otherwise attack any system, whether ours or a third party's.

### 4.4 What is not prohibited

We do not prohibit **lawful sexual content depicting consenting adults**. This
section does not weaken §4.2, which is absolute, and it does not relieve you of
§4.1 — the legality of what you generate, the consent of anyone depicted, your
age-verification and record-keeping obligations, and the licence terms of the
models you select all remain yours.

Stating this explicitly is deliberate. A compute provider that leaves the
question unanswered leaves customers guessing where the line is, and leaves
itself deciding case by case after the fact. The line is §4.2, and it does not
move.

### 4.5 Age

You must be at least 18 years old (or the age of majority where you live, if
higher) to use the Service. If your own product exposes generation capability to
end users, **age verification of those end users is your obligation, not ours**.

### 4.6 Monitoring and enforcement

**We have no obligation to monitor Inputs or Outputs**, and we do not pre-screen
jobs. We may inspect content where necessary to comply with law, respond to a
governmental or judicial request, investigate a report, or protect the Service
and its users.

Where feasible we will notify you of a violation and allow you to correct it. We
may act **immediately and without notice** to throttle, suspend, or terminate
where conduct: (a) exposes us to sanctions, prosecution, or liability; (b) harms
or interferes with the fleet, scheduler, or ledger; (c) interferes with another
customer's use; (d) violates §4.2; or (e) presents an imminent risk of harm.

Suspension for cause stops the plan's rate limits; it does not forfeit purchased
API credits (see §6).

### 4.7 Reporting

Report suspected violations of this section, including CSAM and NCII, to
**abuse@murakumo.cloud**. Include the artifact identifier (CID) or job
identifier where possible. Reports concerning §4.2 are actioned ahead of all
other work.

> [CONFIRM: counsel to review §4 as a whole, and specifically: (a) whether
> §4.4 is consistent with the Operator's payment-processor agreements, given
> that the Service sells compute rather than content; (b) 18 U.S.C. § 2257
> applicability to synthetic outputs and whether a record-keeping obligation
> should be passed through to customers; (c) age-verification duties under the
> UK Online Safety Act, US state AV statutes, and the EU DSA, and whether any
> attach to an infrastructure provider; (d) Japanese Penal Code art. 175 where
> jobs execute on fleet nodes located in Japan; (e) the abuse@ intake and
> NCMEC reporting workflow, including record preservation.]

## 5. Customer data, inputs, and IP ownership

- **Your inputs.** As between you and us, you retain rights in the prompts,
  reference inputs (including referenced content CIDs), and parameters you submit
  ("Inputs").
- **Outputs.** As between you and us, rights in generated artifacts are allocated
  to you or the integrating product's end user, subject to applicable law and to
  the licenses of the models and engines used. [CONFIRM: output ownership and
  any model-license restrictions passed through to you.] Artifacts are stored as
  content-addressed objects; identical content yields identical CIDs.
- **License to operate.** You grant us a limited license to process Inputs,
  execute jobs, produce and store artifacts, and record run-ledger entries solely
  to provide and operate the Service.
- **We do not publish your Outputs.** Artifacts are retained for the account
  that produced them. We do not list, index, feature, or otherwise make them
  publicly discoverable, and we do not use them to market the Service. Retrieval
  requires the account's own credential; a CID alone is not an access grant.
  This is what makes the Service storage-for-you rather than distribution-by-us,
  and it is a design commitment, not merely a current default — see §4.1.
- **Our IP.** We and our licensors retain all rights in the Service, the
  scheduler, executor, run ledger, and related software. Hosted models and
  engines remain subject to their own upstream licenses; nothing here grants you
  rights in them beyond those licenses.

## 6. Fees and billing

- **Metering.** Usage is metered primarily in **GPU-seconds** per run, recorded
  in the run ledger (`:murakumo.run`) together with the model, actor, artifacts,
  and computed cost. The Service also measures throughput (tokens/second) for
  internal cost analysis. [CONFIRM: the actual customer-facing billing unit
  (GPU-seconds and/or tokens) and published prices — do not treat repository
  coefficients, which are marked placeholders, as pricing.]
- **Quotas.** Per-account GPU-seconds limits are enforced fail-closed to prevent
  runaway generation cost.
- **Credits.** Prepaid usage credits are a limited, revocable contractual right
  to consume eligible Service capacity. Credits are not money, stored value,
  property, a bank deposit, or a transferable instrument; have no cash value;
  and may not be redeemed, sold, or exchanged for fiat currency.
- **Payment processor.** Card payments for credits are processed by Stripe.
  Stripe's terms and privacy policy also apply to its processing of payment
  information. We do not receive or store complete card numbers.
- **Credit purchases are final.** To the maximum extent permitted by applicable
  law, all credit purchases are final, non-cancellable, and non-refundable,
  including when credits are unused, partially used, expire, or an account is
  suspended or terminated. Buying credits does not guarantee availability of a
  particular model, engine, region, price, throughput, or Service feature.
- **Required and exceptional refunds.** The preceding no-refund rule does not
  exclude a refund that applicable mandatory law requires. The Operator may
  also, in its sole discretion and without creating a continuing obligation,
  correct a duplicate or unauthorized charge or approve an exceptional refund
  for a verified billing error or material failure by the Operator to provide
  the purchased Service.
- **Effect of a refund, reversal, or chargeback.** If any payment is refunded,
  reversed, disputed, or charged back, we may cancel the corresponding unused
  credits and make an append-only reversal in the credit ledger. If those
  credits have already been consumed or transferred, the account may have a
  negative credit balance. We may suspend paid usage until that balance is
  restored, offset later grants or earnings against it, and recover amounts
  lawfully due. We will not reverse more credits than are attributable to the
  refunded portion of the original purchase.
- **Financial-approval gates.** GPU acquisition and scale-up are treated as
  approval-gated financial side-effects; scale-down is automatic and deletion is
  destructive and approval-gated.
- **Taxes.** Fees are exclusive of taxes unless stated; you are responsible for
  applicable taxes. [CONFIRM the Operator's sales-tax and other collection obligations.]

## 6A. Hardware node purchases (ownership shares)

The store offers two SKUs — *Apple Silicon GPU Node — Ownership Share* and
*NVIDIA GPU Node — Ownership Share* — which are **not** purchases of Service
usage. This section states what they factually are. Until the CONFIRM items
below are resolved by counsel, treat this section as the full extent of what
the Operator asserts about them.

- **What is sold.** Payment is for a physical compute node (Apple Silicon
  Mac mini/Studio class, or a discrete GPU node of l4/h100 class) that the
  Operator procures and deploys into the cloud-murakumo fleet.
- **What the buyer receives.** The node is leased back and operated by the
  Operator, and the buyer receives a share of compute revenue attributable to
  that node under the fleet auction. **Revenue is not guaranteed, not a fixed
  return, and not principal-protected.** Fleet revenue depends on demand,
  auction clearing prices, hardware availability, and the Operator's continued
  operation of the fleet.
- **Fulfillment is manual and not instantaneous.** Unlike credits and plans,
  these SKUs have no automated fulfillment path. Procurement, deployment, and
  the start of any revenue share happen out of band after payment.
  [CONFIRM: delivery/commencement timeline, what happens if the Operator cannot
  procure the hardware, and the refund position in that case.]
- **No equity or governance.** Purchase conveys no equity in the Operator, no
  governance rights, and no claim on the Operator's treasury. This mirrors the
  MCC position stated in §2, and is stated separately here because §2 addresses
  the token and not these SKUs.

> [CONFIRM — counsel review required before these SKUs continue to be sold:
> (a) whether a paid interest in hardware operated by the Operator for a share
> of revenue is an investment contract or other regulated instrument in any
> jurisdiction where the store accepts payment, and if so what registration,
> exemption, disclosure or purchaser-eligibility conditions apply;
> (b) whether selling it through the Operator's payment processor is consistent
> with that processor's restricted-business rules — the store currently accepts
> card payment for these SKUs alongside ordinary usage products;
> (c) what the revenue-share obligation is contractually (term, calculation,
> reporting, termination, transfer, what happens on node failure or fleet
> shutdown); and
> (d) whether §7-§8 disclaimers and the §10 governing-law/dispute provisions are
> appropriate for a purchase of this size and character, or whether separate
> terms are required.
> The Operator has not obtained this review as of the "last updated" date.]

## 6B. Built-to-order node shipments

The store also offers built-to-order machines that are **shipped to the buyer**,
which are different in kind from the §6A ownership shares and are governed by
this section instead. §6A does not apply to them. As of the "last updated" date
these are:

- **murakumo Node 24 — Intel Arc Pro B60 24GB** (¥349,800, Japan only)
- **murakumo Node Edge 8 — Jetson Orin Nano Super 8GB** (¥99,800, Japan only)

Both are sold on the same terms below **except for the warranty term, which
differs per model** (see *Warranty*). Neither carries a performance guarantee
(see *Performance is not warranted*).

*murakumo Node 32 — Intel Arc Pro B70 32GB* was withdrawn on 2026-08-18 and is
no longer offered. Machines already delivered under it remain covered by the
terms in force at the time of their sale.

- **Who sells them.** The statutory Japanese seller disclosure is published
  separately at `/legal/tokushoho`. The public operator of murakumo.cloud is
  Kotoba Labs Inc. Sales contact: Ryo Awai. 法人情報, address, phone and 代表者
  for the hardware seller are disclosed on request. Where this section and
  that disclosure differ, the disclosure governs the sale.
- **What is sold.** A complete, assembled computer that is **shipped to the
  buyer's address** and becomes the buyer's property outright. It is not an
  ownership share, not a lease, and carries no revenue entitlement.
- **Built to order.** The Operator does not hold stock. Payment is taken in full
  at order time; components are then purchased, the machine is assembled and
  tested, and it is shipped. The published lead time is a good-faith estimate,
  not a guarantee. Component prices and availability in 2026 are volatile, and
  the Operator will notify the buyer if the estimate slips.
- **Cancellation and returns.** Cancellation is accepted between payment and the
  point at which components are ordered for that specific machine. After that,
  and after shipment, buyer-initiated cancellation and return are not accepted,
  because the goods are made to order. Defects arising from manufacture or
  transport and reported within 14 days of delivery are repaired or replaced at
  the Operator's cost. Where mandatory consumer law grants a right that this
  paragraph would exclude, that law prevails.
- **Warranty.** From delivery, tracking the component manufacturers' warranties:
  **one year for murakumo Node** (the JPY 99,900 unit currently sold),
  **24 months for murakumo Node 24**, and **12 months for murakumo Node Edge 8**.
  The shorter term on Node Edge 8 is deliberate: its principal component ships as
  a developer kit whose manufacturer warranty the Operator has not confirmed to
  run for 24 months, and the Operator does not state a term it cannot stand
  behind. Modification, disassembly, and external causes (surge, water,
  lightning) are excluded from both.
- **Performance is not warranted.** The published specification (VRAM capacity,
  components) is a description of what is shipped. Inference throughput —
  tokens/second, images/minute, or any comparison between the two models above —
  is **not** part of the specification and is not warranted. The Operator has not
  published measured benchmarks for either machine.
- **Fleet participation is a feature, not a payment.** The machine ships
  configured to serve murakumo fleet inference requests while otherwise idle.
  **This does not entitle the buyer to revenue, credits, or any other
  consideration**, and the Operator does not undertake to send it work.
  Participation is optional, may be disabled by the buyer at any time without
  affecting warranty or support, and electricity is the buyer's cost. If the
  Operator later offers consideration for participation, it will be under a
  separate agreement, not this section.
- **Territory.** Sold and shipped in Japan only. The store rejects checkout for
  this SKU from any other country.
- **Japanese statutory disclosure.** Sale of this SKU is subject to the Act on
  Specified Commercial Transactions. The required disclosure is published at
  `/legal/tokushoho`. **While that disclosure is incomplete the store refuses to
  sell this SKU**, enforced in the storefront code rather than by operational
  undertaking.

> [CONFIRM — counsel review required before this SKU is sold:
> (a) which entity is the importer of record and who bears Japanese customs duty
> and consumption tax, given that the Operator is a Delaware company with no
> recorded Japanese establishment — this determines whether the advertised
> tax-and-shipping-inclusive price is accurate;
> (b) whether selling and shipping goods into Japan creates a Japanese permanent
> establishment, registration, or consumption-tax collection obligation;
> (c) whether the return exclusion above is validly displayed for the purposes of
> Act on Specified Commercial Transactions art. 15-3, and whether the 14-day
> defect window is sufficient under the Civil Code's conformity rules;
> (d) product-safety and electrical requirements applicable to an assembled
> computer supplied in Japan (including PSE treatment of the supplied power
> supply and any labelling obligation on the assembler); and
> (e) whether describing fleet participation as a feature without consideration
> is consistent with how it is marketed elsewhere on the site.
> The Operator has not obtained this review as of the "last updated" date.]

## 6C. Finished mini-PC resale

As of 2026-08-21 the store also offers a finished mini PC resale, which is
different in kind from the §6B built-to-order machines and from the §6A
ownership shares:

- **murakumo Node** (¥99,900, Japan only)

This is a resale of a finished unit. It is not a murakumo-branded built-to-order
node, not an ownership share, and not a fleet-revenue product. The statutory
Japanese seller disclosure is at `/legal/tokushoho`. Shipping address is
collected at Stripe Checkout. A did:key account is not required for this
purchase. Inventory counts and ship dates are not published here.

- **Who sells it.** See `/legal/tokushoho`. The public operator of
  murakumo.cloud is Kotoba Labs Inc. 法人情報, address, phone and 代表者
  for the hardware seller are disclosed on request.
- **What is sold.** One finished mini PC in the 32GB RAM + 512GB SSD
  configuration only. Other memory/storage configurations are not offered.
- **Territory.** Sold and shipped in Japan only.
- **Japanese statutory disclosure.** Sale is subject to the Act on Specified
  Commercial Transactions. The required disclosure is published at
  `/legal/tokushoho`.

## 7. Service levels, warranties, and disclaimers

- The Service is provided **"as is" and "as available"** without warranties of
  any kind, express or implied, including merchantability, fitness for a
  particular purpose, non-infringement, and any warranty as to model output
  accuracy, safety, or suitability, to the maximum extent permitted by law.
- Generative models can produce inaccurate, offensive, or infringing output; you
  are responsible for reviewing and for your use of outputs.
- We do not warrant uninterrupted or error-free operation, committed uptime,
  RTO, or RPO, unless separately agreed in an enterprise contract.

## 8. Limitation of liability

To the maximum extent permitted by applicable law:

- neither party is liable for indirect, incidental, special, consequential, or
  punitive damages, or for lost profits, revenue, data, or goodwill; and
- the Operator's aggregate liability arising out of or relating to the Service
  will not exceed the greater of USD 100 or the fees you paid to the Operator
  for the Service in the 12 months preceding the event giving rise to the
  claim.

Nothing limits liability that cannot be limited under applicable mandatory law.

## 9. Term and termination

- These Terms apply while you use the Service.
- You may stop submitting jobs at any time.
- We may suspend or terminate access for material breach, non-payment, legal
  requirement, cost-control risk, or risk to the fleet, with notice where
  practicable.
- On termination we may delete account and run-ledger data subject to legal
  retention obligations and to the content-addressing limits described in the
  Privacy Policy.
- Ownership, disclaimers, limitation of liability, and governing-law provisions
  survive termination.

## 10. Governing law and dispute resolution

- These Terms are governed by Delaware law, without regard to its
  conflict-of-laws rules.
- Except where mandatory consumer law permits proceedings elsewhere, the state
  courts located in New Castle County, Delaware and the United States District
  Court for the District of Delaware have exclusive jurisdiction. Each party
  consents to personal jurisdiction and venue there.
- These Terms do not require arbitration and do not waive a right to a jury
  trial.
- Nothing in this section deprives a consumer of non-waivable protections or a
  forum available under mandatory law in the consumer's habitual residence.

## 11. Changes

We may update these Terms; material changes are indicated by updating the "Last
updated" date and, where appropriate, by additional notice. [CONFIRM notice
mechanism.] Continued use after changes take effect constitutes acceptance.

## 12. Contact

Customer and legal notice email: support@murakumo.cloud, designated by the Operator
as its current electronic notice channel.

- Site / service operator: Kotoba Labs Inc
- Sales contact: Ryo Awai
- 法人情報 / address / phone / 代表者: disclosed on request at
  support@murakumo.cloud. This repository does not contain an Inc filing
  for Kotoba Labs Inc; those values are not invented here.