Peer-to-Peer Network Processing Addendum — 1.2 (archived)

Version 1.2 · Dated: 28 October 2026 · Version archive

This addendum supplements the Terms of Service of Decision Models, provided by productivity-boost.com Betriebs UG (haftungsbeschränkt) & Co. KG ("we", "us"). It applies only to API keys for which Peer-to-peer network processing is enabled. Network processing is set per key in the dashboard (see section 5). For Peer-to-Peer keys created on or after 28 October 2026 it is preselected; the customer can choose "Our GPUs only" before creating the key and can switch it off at any time. EU-tier keys never use it. It does not require, and does not enable, the key's Allow worldwide processing setting, which is a separate switch. Keys created before 28 October 2026 are not changed: their switch stays as it was, and network processing is enabled for them only by an active choice of the customer. EU-tier requests never use the network, whatever the key's settings. Allowing worldwide processing alone does not enable network processing.

1. Scope and providers

Network processing is available for keys on which it is enabled. Eligible requests are single-question text requests in the Peer-to-Peer tier to the exact model versions the network serves: s1-fast-1 at launch (the dashboard shows the current list from the public network status). On such keys they are sent to the peer-to-peer network of independent compute providers as the primary route, with our own GPUs as fallback. If the network has no capacity for an exact network version, the request is answered on our own GPUs with the same exact model version, never with another model. A retryable 503 is returned only if neither the network nor our own GPUs have capacity. All other requests, and all requests using keys with network processing off, are processed without the peer-to-peer network, as described in the Terms. EU-tier requests are processed only in the EU, never by the peer-to-peer network, with no payload retention.

The independent providers are known as "miners" in the REAX network, which runs on Bittensor subnet 92. They follow the network's protocol rules and their answers are checked as described below. They are anonymous third parties located worldwide and outside our control. They are not bound to us by contract, in particular not to confidentiality or to deleting content.

2. Verification, fallback and price

Network answers are checked on a sampled basis against our own reference GPU running the same model before they are returned, and every new provider is checked on its first answers. Answers not selected for a check are returned without a comparison against our reference GPU. If a provider is unavailable or slow, or a checked answer deviates, our own GPU answers instead. The normal Peer-to-Peer-tier price applies. Checking concerns the answer; it does not establish confidentiality or prevent a provider from storing content.

3. Content exposure and retention

Providers receive the request content — state, question and options — in clear text to compute the answer. We cannot technically prevent a provider from storing what it sees. No confidentiality or deletion guarantee applies to content received by providers. We send no account, API key, IP or original request identifiers to them. Identifiers the customer includes in the content itself remain visible to them.

We do not store request or response payloads. Our no-retention commitments apply to our own infrastructure and do not extend to independent network providers. The sub-processor list describes this network processing and its recipients.

4. Customer obligations

Do not send personal data, secrets or confidential content on keys with network processing enabled. Use the EU tier for such content. Enabling either processing switch does not supply a legal basis for transferring personal data. The customer is responsible for ensuring that content sent on network-enabled keys meets these restrictions.

Relationship to the DPA: By creating or keeping a key with network processing enabled (including by leaving the preselected network option when creating the key), the customer instructs us to make eligible request content of that key available to the network providers. Because no personal data may be sent on network-enabled keys, network providers are not sub-processors under section 6 of the DPA and are not listed in its Annex 3; the DPA's sub-processor, notice and third-country provisions (sections 6 and 7) are not applied to them, and no Art. 28(4) GDPR contract, Standard Contractual Clauses or other safeguard under Chapter V GDPR exists or can be put in place with them. Statements in our Terms, DPA, Privacy Policy and data-retention or security pages that content is not stored 'on any node' or 'on every node' refer to infrastructure we operate or rent, not to network providers.

5. Enabling and switching off

Network processing is set per key in one of two ways: (a) when creating a new Peer-to-Peer key in the dashboard, the customer sees the choice between "Peer-to-peer network first for supported models" and "Our GPUs only" together with a summary of this addendum and the notice not to send personal data on network-enabled keys; for keys created on or after 28 October 2026 the network option is preselected, and the customer can select "Our GPUs only" before creating the key; or (b) in the key list, where the dashboard displays a summary of this addendum before the customer confirms enabling it for an existing key. Keys created before 28 October 2026 are never switched on implicitly. For a Peer-to-Peer key created through the API on or after 28 October 2026, network processing follows the request field allow_p2p_subnet_processing; if the request does not set it, network processing is on. The customer can switch it off at any time. Choosing network processing enables only network processing; it does not turn on the key's Allow worldwide processing setting. Each key's current setting is shown in the dashboard key list. The customer may switch network processing (and, separately, worldwide processing) off at any time, effective for new requests. Switching it off does not cancel requests already admitted or undo content already disclosed to a provider.

This addendum does not change the EU tier, keys created before this version took effect (unless the customer enables network processing for them), or keys whose network-processing switch is off, including keys that allow worldwide processing without network processing. The worldwide-processing setting and its conditions remain separate and are not affected by this addendum.

6. Relationship to the Terms

For enabled keys only, this addendum describes the additional processing recipients and the limits of retention and confidentiality commitments for those recipients. All other provisions of the Terms, including liability and warranty, remain unchanged. This addendum creates no new warranty or guarantee and does not exclude or limit liability beyond the Terms. German version.