API Data Policy
1. Scope
This Data Policy explains how AccelerAI and its affiliates, including its subsidiary MoBagel, Inc. ("AccelerAI," "we," "us"), handle content sent to the AccelerAI API, whether you call it directly or through an aggregator such as OpenRouter (the "API").
It forms part of our Terms of Service, which cover Customer Data in our products, and supplements your agreement with us. Our Privacy Policy covers only personal data collected through our website. This Data Policy does not cover other AccelerAI products, on-premises deployments, or information you send us outside the API, such as through our contact form or support.
2. Definitions
“Customer Content” means everything you send to the API (prompts, messages, system instructions, files, images, tool definitions and tool results) and everything the API returns to you. Customer Content is Customer Data under our Terms of Service.
“Intermediate Content” means anything the API derives from Customer Content while serving a request, including sub-prompts, chunks, routing decisions, and the inputs and outputs of any model the API coordinates. Intermediate Content is Customer Content for every purpose in this policy.
“Zero Data Retention” ("ZDR") means Customer Content is processed only in memory, only as needed to serve the request, is not logged, and is not kept after the request completes.
“Operational Metadata” means the data in section 6, which does not include Customer Content.
For Customer Content, AccelerAI acts as a data processor on your behalf, and you are the data controller. For your account and billing information, AccelerAI is the data controller.
3. Our commitments at a glance
- ZDR applies to all API traffic. You do not need to enable it or sign anything.
- We never use Customer Content to train, fine-tune, evaluate, benchmark or otherwise improve any model or system, including our routing and orchestration models.
- We never sell or share Customer Content with advertisers, data brokers or model developers, except a third-party model you choose to use, as described in section 8.
- You own your inputs and outputs.
4. How we process Customer Content
The API breaks a request into parts and coordinates several open-weight models to answer it. All of this happens in memory on infrastructure we control or rent. Customer Content and Intermediate Content are not written to disk, databases, logs or traces at any step. When the response is returned, they are discarded.
Caching. To make repeat requests faster and cheaper, we may hold prompt and key-value caches in GPU or accelerator memory. Caches hold model computations, not text, are used only to serve your requests, are never shared across accounts, and are overwritten as the cache is reused. Caching is not retention under this policy.
5. No training or secondary use
We do not use Customer Content or Intermediate Content to train, fine-tune, distill, evaluate, benchmark or improve any model, including AccelerAI models, our routing, chunking or orchestration logic, and any draft or speculative-decoding model. We improve the API only with Operational Metadata, public data, synthetic data and data customers give us under a separate written agreement that names this purpose.
6. What we keep
We keep the following Operational Metadata for each request: account and API key identifiers; the model and API version requested; timestamps; input, output and cached token counts; latency; status and error codes; and serving region. If you call us through an aggregator, we receive that aggregator's request ID and not your identity unless the aggregator passes it.
We keep Operational Metadata for 90 days for reliability, security and capacity planning, and keep billing records as long as tax and accounting law requires.
7. Where data is processed
We serve requests from United States data centers. Requests that reach us through OpenRouter are served from the United States only. If we add a serving region, we will name it here at least 30 days before it serves your traffic, and enterprise customers may pin their traffic to a region.
8. Service providers and third-party models
We run inference on cloud and hardware providers that process data only on our instructions, as our data processors, under written confidentiality, security and data-protection terms. The current list may be requested at any time at contact@acceler.ai. We will give 30 days' notice before adding one, and enterprise customers may object on reasonable grounds.
The AccelerAI API on OpenRouter uses only open-weight models we host; no Customer Content is sent to third-party model developers. On our direct API, if you choose to bring your own key for a third-party model, the content sent to that model is governed by your agreement with its provider.
9. Safety and abuse
We may screen requests in memory, as they are processed, to detect abuse and illegal content. Screening does not create logs of Customer Content. If content is flagged as potential child sexual abuse material, we may keep it only for the review, reporting and preservation that law requires. We will not keep content for general abuse investigation.
10. Security
We encrypt traffic in transit with TLS 1.2 or higher, encrypt stored Operational Metadata at rest, limit internal access by role, and keep audit logs of that access.
11. Legal requests
Because we do not keep Customer Content, we generally cannot produce it later, including for support or legal process. If a binding legal order requires us to preserve or disclose content from future requests, we will do so only to the extent required and will notify the affected customer first, where we can identify them, unless the law forbids it.
12. Your rights
If you call us through an aggregator, we hold no account data about you; contact the aggregator for account matters. Account holders may request access to, correction of, restriction of, export of or deletion of their account data and Operational Metadata at contact@acceler.ai. Requests are free of charge and answered as early as possible, and always within one month. If you have an account with us, when you close it we delete associated Operational Metadata within 30 days, except billing records the law requires us to keep. You may also lodge a complaint with your data protection authority. For transfers out of the EU, EEA or UK, we use the European Commission's standard contractual clauses. A Data Processing Agreement is available on request.
13. Order of precedence
For Customer Content sent to the API, where this Data Policy conflicts with our Terms of Service, our Privacy Policy or any license in them, this Data Policy controls. A signed agreement can change it only by naming the sections it changes.
14. Changes
We will post changes on this page. For material changes we will email account owners and post a notice at least 30 days before the change applies. Changes apply only to requests received after they take effect.
15. Contact
AccelerAI and its affiliates, including its subsidiary MoBagel, Inc.. 3031 Tisch Way, 110 Plaza West, San Jose, CA 95128, USA.
contact@acceler.ai
