AI Usage Policy

AI Services We Use

Sanary (the “Services”) uses artificial intelligence in certain features. Depending on the application, device, and feature, Sanary primarily uses:

• AI features using Apple Private Cloud Compute (PCC); and
• Anthropic Claude through Amazon Bedrock.

Not every application or feature uses every AI service listed above; the AI services used differ by application and feature.

Sanary Build provides “APCC (Adaptive Production Creator Core),” an AI agent feature using Apple PCC. Another AI service may be used depending on availability and other operating conditions.

If Sanary adds an AI service or makes a material change to its use of AI services, Sanary will update this Policy and provide notice.

Purposes of AI Features

Within the scope identified in each application, Sanary uses AI for:

• record entry, organization, and reflection;
• conversation, text generation, and input assistance;
• investigation of projects and files;
• creation, editing, and review of source code and similar materials; and
• other AI-assisted features identified in the applicable application.

AI output is provided for reference and is not guaranteed to be accurate or complete.

Data Sent to and Processed by AI Services

When you use an AI feature, input and other information necessary to process your request may be processed by the AI service used for that feature.

For record-oriented applications, Sanary may remove or de-identify direct identifiers where the feature allows and doing so is appropriate. Sanary does not, however, represent that all AI features apply de-identification in every case.

In Sanary Build, project information, file contents, source code, instructions, and similar information may be processed to the extent necessary to perform work you request. Sanary Build does not collect or store your projects or source code on Sanary servers, but that does not mean no external processing occurs when you use an AI feature.

Risk of Inaccurate Output

AI may produce information that is inaccurate, outdated, or inappropriate for the context. Verify output before relying on it for an important decision.

AI may not accurately understand your circumstances or intent.

Source code, file changes, commands, and other development assistance may cause unintended changes, vulnerabilities, or data loss. Review proposed changes and perform appropriate testing and backups before applying them.

Model Training and Data Retention

The use and retention of AI inputs and outputs depend on the service, model, feature, and applicable contractual terms. As of August 31, 2026, the principal conditions confirmed by Sanary are:

• Apple PCC: Apple states that user data sent to PCC is used solely to fulfill the request, is not retained after the response is returned, and is not accessible to Apple personnel. Accordingly, user data processed by PCC is not used for model training. See Apple’s official explanation.
• Anthropic Claude through Amazon Bedrock: Some Sanary applications and features use Claude through Amazon Bedrock. AWS states that Bedrock inputs and outputs are not used to train the underlying models and generally are not shared with third-party model providers. Exceptions concerning retention may apply depending on the model, feature, legal obligations, or abuse prevention. See the Amazon Bedrock FAQ and AWS data-retention documentation.
• Anthropic commercial services: Anthropic states that inputs and outputs from its commercial services are not used for model training unless the user expressly permits that use or another stated exception applies. Anthropic also states that direct API inputs and outputs are generally deleted within 30 days. Where Sanary uses Claude, it does so through Amazon Bedrock—not the direct API or consumer Claude—so that 30-day period is not stated as a retention rule for Sanary’s Bedrock use. See Anthropic’s official explanations of model training and retention.

Sanary does not use user data sent through AI features to train Sanary’s own AI models. Sanary will update this Policy if the terms of an external service change materially.

Responsible Use

Do not submit personal information, secrets, credentials, third-party confidential information, or other information that is not necessary for the AI feature. Before processing a project or source code, confirm that you have lawful authority to process that information.

Changes to AI Features

The content, accuracy, and availability of AI features may change without notice as AI services and provider policies evolve.

Revision History

• September 19, 2026: Clarified that the AI services used differ by application and feature, and revised the statements on removing or de-identifying identifiers and on the use of Anthropic Claude through Amazon Bedrock to reflect that they vary by application and feature.
• August 31, 2026: Added Sanary Build, Apple Private Cloud Compute, and Anthropic Claude through Amazon Bedrock, and revised the provisions on purposes, processed data, training, and retention.
• August 8, 2026: First published as a common AI Usage Policy, separating and consolidating common provisions from the Sanary Log AI policy.