Provider Terms

2.1 Introduction

This Provider Agreement ("Agreement") is a legally binding contract between you ("Supplier," "you," or "your") and dotgear, inc ("InfrPool," "Platform," "we," "us," or "our"). This Agreement governs your participation as a computing resource supplier on the InfrPool platform.

This Agreement supplements the InfrPool Terms of Service ("ToS"), which are incorporated by reference. In the event of a conflict between this Agreement and the ToS, this Agreement prevails with respect to Supplier-specific matters.

By registering as a Supplier, activating Worker software, or listing a model on the Platform, you acknowledge that you have read, understood, and agree to be bound by this Agreement.

2.2 Supplier Role and Relationship

Role Definition

As a Supplier, you provide computing resources to the Platform for the purpose of fulfilling AI inference requests from Consumers. You may provide resources through one or both of the following supply types:

• Self-Hosted Models: You operate your own hardware (GPUs, servers) running open-source or self-hosted AI models, and connect them to the Platform via the InfrPool Worker software.
• CLI Proxy: You make available your Claude Code CLI subscription capacity to fulfill Claude model requests routed through the Platform.

Independent Relationship

You are an independent resource provider, not an employee, agent, partner, or joint venturer of InfrPool. Nothing in this Agreement creates an employment, agency, or partnership relationship. You are solely responsible for your own taxes, insurance, and compliance with applicable laws.

Responsibility Model

InfrPool operates a marketplace responsibility model. As a Supplier:

• You are solely responsible for the outputs generated by your models. InfrPool is a matchmaking platform and does not control, inspect, or moderate the content of model outputs.
• You are responsible for ensuring that your computing resources operate in compliance with all applicable laws and regulations in your jurisdiction.
• You are responsible for the configuration, maintenance, and security of your own hardware and software.
• If a third-party claim arises from content generated by your model, you bear sole responsibility for that claim.

2.3 Model Quality and Benchmarks

Initial Verification

Before your model is listed on the marketplace, it must pass an initial benchmark conducted by the Platform:

Periodic Re-Verification

Your models are subject to periodic re-verification every 24 hours (lightweight version: 3 knowledge questions + 2 instruction tasks + speed measurement).

• If your model fails re-verification, it is automatically taken offline and you are notified.
• A re-test is attempted 1 hour after failure.
• Three consecutive failures result in removal of the model from the marketplace.

Quality Score

Each model receives a quality score (0-100) based on:

• General knowledge accuracy (40% weight)
• Instruction following accuracy (40% weight)
• Latency score (20% weight)

Anti-Cheating Measures

Benchmark questions are drawn randomly from a server-side pool that is not disclosed to Suppliers. The pool is refreshed monthly (10-20% replacement). If a model's quality score increases suddenly by more than 15 points, a full re-verification (50 questions) is triggered automatically.

2.4 Pricing and Fees

Price Setting

You set your own prices for your models. For text inference, you set separate input and output token prices (per 1M tokens, in micro-USD). For image generation, you may set per-size pricing. There is a minimum base price of $0.000001 per 1M tokens (1 micro-USD) to ensure non-zero earnings.

Supplier Fee (Rank-Based)

Your earning per request is calculated as: base_price x (1 - fee_rate), rounded down (floor) to the nearest micro-USD.

2.5 Rank System

If your 30-day rolling success rate falls below the maintenance threshold for your current rank, you are demoted by one rank in the daily batch process (UTC 04:00). After demotion, a 7-day cooldown applies before you are eligible for re-promotion.

Success Rate Calculation

Success Rate = Successful Requests / (Successful + Failed) x 100%

• Successful: Normal response delivered to Consumer.
• Failed: Timeout, error, or disconnection.
• Neutral (not counted): Provider-side 429, Consumer cancellation, self-consumption.

2.6 Payouts

All Supplier payouts are processed exclusively through the Tremendous API. Available payout options include Visa prepaid cards, PayPal, Amazon gift cards, bank transfers, and 1,000+ other brands supported by Tremendous.

Your wallet balance (including earnings) expires one (1) year after the date of last activity. A 30-day advance warning email is sent before expiry. Expired credits are non-refundable.

2.7 Prompt Data Obligations

You must not log, store, record, cache, or persist in any form the content of Consumer prompts or model responses beyond what is strictly necessary for processing the immediate inference request in real time.

You must not share, sell, transmit, publish, or otherwise redistribute any Consumer prompt or response data to any third party, for any purpose, including but not limited to model training, analytics, research, or commercial use.

Your Worker software must process prompts in memory only. Upon completion of each inference request, all prompt and response data must be discarded from your system's memory.

Violation of prompt data obligations constitutes a material breach of this Agreement and will result in immediate and permanent ban from the Platform, forfeiture of all pending wallet balance, and potential legal action.

Reference Image and Generative Content Obligations

If you provide image generation or any model capable of processing reference images:

• You must not log, store, retain, or cache any reference images, uploaded images, or generated outputs beyond the immediate processing period.
• You must not use reference images or generated outputs for model training, fine-tuning, or dataset construction.
• You must cooperate with the Platform in responding to takedown requests, law enforcement inquiries, or abuse reports.
• You must implement reasonable technical safeguards to prevent unauthorized retention of reference image data.

2.8 CLI Proxy — Self-Responsibility Clause

Nature of CLI Proxy

CLI Proxy is a supply method in which you make your Claude Code CLI subscription capacity available to fulfill Claude model inference requests routed through the Platform.

Your Sole Responsibility

By enabling CLI Proxy, you acknowledge and accept the following:

• The use of your Claude Code CLI subscription to fulfill third-party inference requests may violate Anthropic's Terms of Service.
• You assume all risk associated with using your Anthropic subscription in this manner, including account suspension or termination by Anthropic.
• InfrPool makes no representation that CLI Proxy usage is permitted under Anthropic's Terms of Service.
• Any dispute between you and Anthropic arising from your use of CLI Proxy is solely your responsibility.

Kill Switch

InfrPool reserves the right to instantly and unilaterally disable all CLI Proxy functionality across the entire Platform at any time, without prior notice. You acknowledge that the kill switch may be activated at any time and agree that InfrPool shall have no liability for any loss of earnings or other damages resulting from kill switch activation.

CLI Proxy Is Not the Core Service

CLI Proxy is positioned as a supplementary supply method alongside self-hosted models. If CLI Proxy is permanently disabled, the Platform will continue to operate with self-hosted model supply only.

2.9 Fraud and Prohibited Conduct

The following actions constitute fraud and are strictly prohibited:

• Fake models: Registering a model that misrepresents its identity.
• Token inflation: Artificially inflating token counts to increase earnings.
• Response tampering: Modifying, degrading, or replacing model responses in transit.
• Prompt theft: Logging, storing, or redistributing Consumer prompts.
• Benchmark manipulation: Attempting to game benchmark tests.
• Self-consumption abuse: Using multiple accounts to artificially inflate transaction volume.
• Rank manipulation: Creating multiple accounts to gain unfair ranking advantages.

Confirmed fraud results in immediate and permanent ban from the Platform, forfeiture of entire wallet balance, and potential reporting to relevant authorities.

2.10 Removal and Suspension

Your models may be automatically suspended if:

• Three consecutive benchmark failures occur.
• Token count deviation exceeds 10% on 3 consecutive requests.
• Five consecutive request errors occur.
• Your success rate drops below rank maintenance thresholds.

Upon suspension:

• All models are taken offline and removed from the matching pool.
• Pending earnings are frozen.
• API and dashboard access remains available in read-only mode.
• You will be notified by email with the reason for suspension.

2.11 Intellectual Property

You represent and warrant that you have all necessary rights, licenses, and permissions to operate the AI models you make available through the Platform, including compliance with any applicable open-source licenses.

The InfrPool Worker software, APIs, and related tools are provided to you under a limited, non-exclusive, non-transferable, revocable license solely for the purpose of participating as a Supplier on the Platform. You may not reverse-engineer, modify, or redistribute the Worker software.

2.12 Indemnification and Limitation of Liability

You agree to indemnify, defend, and hold harmless InfrPool from and against any and all claims arising out of or in connection with:

• The content generated by your models.
• Your violation of this Agreement or the Terms of Service.
• Your violation of any applicable law or third-party rights (including Anthropic's Terms of Service if using CLI Proxy).
• Any third-party claim related to your computing resources or model outputs.
• Your failure to comply with prompt data obligations.

TO THE MAXIMUM EXTENT PERMITTED BY APPLICABLE LAW, INFRPOOL SHALL NOT BE LIABLE FOR ANY INDIRECT, INCIDENTAL, SPECIAL, CONSEQUENTIAL, OR PUNITIVE DAMAGES ARISING OUT OF YOUR PARTICIPATION AS A SUPPLIER. INFRPOOL'S TOTAL AGGREGATE LIABILITY UNDER THIS AGREEMENT SHALL NOT EXCEED THE TOTAL AMOUNT OF PLATFORM FEES COLLECTED FROM YOUR TRANSACTIONS IN THE TWELVE (12) MONTHS PRECEDING THE EVENT GIVING RISE TO THE CLAIM.

2.13 Term, Termination, and Governing Law

This Agreement is effective from the date you register as a Supplier and continues until terminated by either party. You may terminate by withdrawing your earnings, removing all models, and requesting account deletion. InfrPool may terminate immediately if you breach any provision.

Sections covering Prompt Data Obligations, CLI Proxy Self-Responsibility, Consequences of Fraud, Indemnification, and Limitation of Liability survive termination of this Agreement.

This Agreement shall be governed by and construed in accordance with the laws of Japan. Any disputes shall be resolved in accordance with the dispute resolution provisions in the Terms of Service.

For questions regarding this Agreement, contact us at: support.infrpool@dotgear.jp