# Data and privacy

> LMU AI data handling: request and response content is never persisted, call metadata is kept for 90 days, and account data is stored only within China.

URL: https://docs.lmuai.ai/docs/guide/privacy



This page explains the principles LMU AI follows when handling user data. We treat "minimal storage, isolation, and transparency" as the three baselines of our data handling.

## What data we handle [#what-data-we-handle]

We divide data into two categories: **account data** and **request data**. The two are handled and stored in completely different ways.

### Account data [#account-data]

| Type                   | Content                                   | Purpose                                 |
| ---------------------- | ----------------------------------------- | --------------------------------------- |
| Identity information   | Email, phone number                       | Login, password recovery, notifications |
| Billing information    | Plan, quota, order records                | Billing and invoices                    |
| Credential information | API key (stored encrypted), login session | Authentication                          |
| Usage metadata         | Call time, model, token usage, duration   | Billing, statistics, reports            |

### Request data [#request-data]

| Type          | Content                                       | Purpose                                                |
| ------------- | --------------------------------------------- | ------------------------------------------------------ |
| Request body  | The prompt, messages, and parameters you send | Forwarded to the upstream model to generate a response |
| Response body | The content the model returns                 | Returned to you                                        |

**Key distinction**: account data is used to operate the platform itself; request data is used only to complete your current, single API call.

## Where data is stored [#where-data-is-stored]

Our infrastructure has two layers:

**Access layer (within China)**

* Main servers, user database, billing system
* **All account data is stored here**
* Your email, phone number, orders, and API key never leave China

**Upstream channel layer (multi-region, including overseas)**

* Used to access the official interfaces of model providers such as OpenAI / Anthropic / etc.
* These channels **serve only as a relay for model traffic** and do not persist any user identity information
* Channel nodes do not store user accounts and do not know which user a request comes from (they see only the anonymized request payload)



## Log retention policy [#log-retention-policy]

| Data type                                                 | Retention period                                                    | Storage location |
| --------------------------------------------------------- | ------------------------------------------------------------------- | ---------------- |
| Request / response body content                           | **Not persisted**                                                   | —                |
| Call metadata (time, model, token, duration, status code) | 90 days                                                             | Within China     |
| Error logs (may include truncated request fragments)      | 90 days                                                             | Within China     |
| Account and order data                                    | For the life of the account; cleared within 30 days of cancellation | Within China     |

**About request content**

* We **do not store your prompt or response content** — once a request is forwarded, it is released from memory and never written to any database or file
* When an upstream channel returns an error, the error log may contain **truncated request fragments** (used to reproduce the problem); this data is cleared after 90 days
* Error logs are not used to train any model and are not proactively reviewed by humans; they are only accessed by the relevant engineers when you submit a ticket that needs troubleshooting help

<Callout type="info" title="Optional audit logs for the Enterprise tier">
  Enterprise customers can opt in to the audit log feature and retain request / response data according to your company's compliance requirements. Once enabled, logs are stored encrypted, and access can be controlled by your company's administrators. See [Enterprise plans](/docs/enterprise) for details.
</Callout>

## Cross-border data transfer [#cross-border-data-transfer]

Because the servers for upstream models (Claude, GPT, etc.) are located overseas, **the request and response bodies of an API call pass through overseas channels during forwarding**. This is a necessary technical path for any relay service that accesses overseas large-model APIs.

Our commitments at this stage:

* **We don't persist any data on overseas nodes**: the channels only forward, and memory is released as soon as forwarding completes
* **We don't carry user identity**: the requests the upstream channels see have already been stripped of account fields such as email, phone number, and order associations
* **Encrypted transmission**: user ↔ access layer, access layer ↔ upstream channel, and upstream channel ↔ model provider are all TLS 1.2+ end to end
* **We don't share with third parties**: apart from the upstream model providers necessary to complete the current call, we don't pass your request content to any third party

<Callout type="warn" title="Advice on sensitive data">
  Because request content must pass through overseas channels to reach the model provider, **do not send requests through the API that contain highly sensitive data such as state secrets, trade secrets, or personal ID numbers**. If you have such compliance requirements, contact sales to discuss on-premises deployment or a compliant channel.
</Callout>

## What we don't do [#what-we-dont-do]

* **We don't read** your prompt or response content for human analysis, commercial analysis, or any model training
* **We don't sell** any of your data to third parties
* **We don't retain** request and response content **without informing you**
* **We don't transfer account data across borders** (account data is stored only within China)

## Security practices [#security-practices]

* API keys are stored encrypted in the database, showing only a few leading and trailing characters
* All internal operations that access the database are audit-logged
* Key changes (plan, API key reset, withdrawals, etc.) trigger email / SMS notifications
* Team members accessing production data must go through approval and the principle of least privilege

If you notice a suspected account anomaly, please immediately:

1. Reset your API key in the console
2. Contact sales or support: [business@lmuai.com](mailto:business@lmuai.com) / WeChat 18599001010

## Enterprise compliance support [#enterprise-compliance-support]

Enterprise customers with any of the following needs should reach out through the business channel:

* Signing a Data Processing Agreement (DPA)
* Custom log retention policies
* On-premises gateway deployment (request content does not pass through shared channels)
* Supporting materials for MLPS / GDPR / SOC2 questionnaires

## Policy changes [#policy-changes]

If there are material changes to this policy, we will post a notice in a prominent place on the website and notify active users by email. Continuing to use the service after a policy update constitutes acceptance of the new policy.

*Last updated: 2026-05-06*