---
title: "Zero data retention LLM API: what is and isn't stored"
description: "BroadNet Inference does not retain prompts, images, audio, answers or transcripts. What it does keep, how the hourly check works, and where the GPUs run."
url: https://broadrouter.com/zero-data-retention/
lang: en
updated: 2026-11-15
publisher: BroadRouter (a BroadNet site); API by BroadNet Inference, https://inference.broadnet.ai
---

# Zero data retention at BroadNet Inference

Updated 2026-11-15

BroadNet Inference processes your prompts, images, audio, answers and transcripts for the request and does not retain them. It keeps only content-free records such as timings, token counts and cost, and does not train on your data. BroadNet checks its gateway configuration hourly. Hosted models run on BroadNet-owned or leased US hardware; partner-served models run under zero-data-retention terms.

BroadNet's docs describe zero data retention as "built into how we serve requests, not only stated in policy". This page summarises that statement. The full list of fields is in the [privacy statement](https://www.broadnet.ai/inference-privacy).

What is kept, and what is not

1.  **Inference content**
    
    -   Prompts and messages
    -   Images
    -   Audio
    -   Answers and transcripts
    
2.  **The model answers**Content is used to serve your request
    
3.  **Not retained**Not used to train, fine-tune or evaluate models
    

**Retained: content-free records**

-   Account
-   Billing
-   Traffic
-   Security

For operational and legal requirements

Inference content is not retained. Necessary account, billing and security metadata is retained as described in our privacy statement.

[Inference privacy statement](https://www.broadnet.ai/inference-privacy)

## What is never retained?

-   Prompts, images and audio you send
-   Answers and transcripts the models return
-   Request and response bodies
-   Raw URLs and query strings
-   Raw authentication values and session cookies
-   Arbitrary upstream metadata

BroadNet also says it does not write inference content to logs, including error logs, and does not use your inference data to train, fine-tune or evaluate models.

## What does BroadNet keep?

| Record | Examples | Why |
| --- | --- | --- |
| Traffic records | Request IDs, times, route, HTTP status, byte counts, timings, client IP | Security and operations |
| Usage accounting | Account and key IDs, model, token counts, cost, cache hit | Billing |
| Account and billing | Email, key status, SHA-256 hashes of keys, allowed IPs, balances, payments | Running your account |

## Why it matters for agents

_[image: Documents passing through a glass capsule and leaving it, beside the text "Zero content retention.", "BroadRouter" and "Powered by BroadNet Inference".]_

Illustration. Inference content is not retained; content-free account, billing and security records are.

Coding agents send whole repositories, and support bots send customer messages. With zero data retention, that content is not retained once the request finishes. Your [keys can also be locked to your own servers' IP addresses](https://broadrouter.com/guides/api-keys-ip-allow-lists/).

## How is retention checked?

BroadNet says its gateway configuration is checked every hour for content-free operation; its docs call this an automated retention audit. It is BroadNet's own check. BroadNet does not publish a third-party audit report for the Inference API, so ask for one if your compliance team needs it. The statement also promises that BroadNet will update the page before any change that weakens these retention guarantees.

## Where does processing happen?

Per BroadNet's privacy statement, its own models and the open-weight models it hosts run on BroadNet-owned or leased hardware in the United States, with partner capacity as backup; partner-served models run under zero-data-retention terms with US inference partners, and a subprocessor list is available to customers on request. BroadNet's Terms of Service name DeepSeek V4.1 Flash as partner-hosted. For data-residency needs, agree processing locations with BroadNet before you send data; its inference page says regional processing in MENA or Europe is available on request, and private endpoints run in the region you need. [More about BroadNet](https://broadrouter.com/about/).

## Does it cover access through other platforms?

Partly. The statement covers inference.broadnet.ai and API access sold through third-party platforms, but those platforms handle your data under their own policies before it reaches BroadNet. If you call the models through a marketplace or a chat app, check that platform's retention terms too.

## Questions

### Is my data used to train models?

No. BroadNet's privacy statement says it does not use your inference data to train, fine-tune or evaluate models, and prompts, images, audio, answers and transcripts are not retained after the request.

### What does BroadNet keep?

Content-free records: timings, token counts, cost, status codes, IP addresses, and account and key identifiers. They are kept as long as billing, tax, security and legal obligations require.

### How is it checked?

BroadNet says its gateway configuration is checked every hour for content-free operation. It describes this as an automated retention audit. It is a check on configuration, run by BroadNet; BroadNet does not publish a third-party audit of it.

### Does it apply when I use the models through another platform?

BroadNet's statement applies to inference.broadnet.ai and to API access sold through third-party platforms, but those platforms process your data under their own policies before it reaches BroadNet. BroadNet says the same models are on Poe; Poe's own terms apply to what Poe keeps.

### Is the connection encrypted?

Yes. The privacy statement says the API is served over HTTPS and traffic between BroadNet's internet edge and its servers is encrypted in transit. Scoped API keys control access by model, rate and spend, and keys can be locked to your IP addresses.

### Where are the GPUs?

BroadNet's privacy statement says its own models and the open-weight models it hosts run on BroadNet-owned or leased hardware in the United States, with partner capacity as backup. Partner-served models, such as DeepSeek V4.1 Flash, run under zero-data-retention terms with US inference partners. Processing locations for residency needs are agreed with BroadNet before you send data; BroadNet says regional processing in MENA or Europe is available on request.

Try it on your real workload. New accounts start with $5 of credit for Gemma and transcription. DeepSeek, GLM and Qwen3.8 need a top-up.

[Get started](https://inference.broadnet.ai/account/?utm_source=broadrouter&utm_medium=referral&utm_campaign=zero-data-retention&utm_content=page-cta) ·

**Sources**

-   [BroadNet Inference privacy statement, effective September 2026 (checked 2026-11-15)](https://www.broadnet.ai/inference-privacy)
-   [API reference: where your data goes](https://inference.broadnet.ai/docs?utm_source=broadrouter&utm_medium=referral&utm_campaign=zero-data-retention&utm_content=sources#data)
-   [BroadNet Inference overview: data handling and FAQ (checked 2026-11-15)](https://www.broadnet.ai/inference)
-   [BroadNet Terms of Service, §7.3 hosting (checked 2026-11-15)](https://www.broadnet.ai/terms)
