saidosi

Privacy Policy

This policy explains how saidosi handles the data needed to provide accounts, AI, sync, reliability, and support. We do not sell personal information, show advertising, or track you across apps.

Updated: November 2, 2026

Scope and contact

This policy applies to the saidosi iOS app and its supporting services. The personal information processor is ZHIYI LIN ("we"). To exercise your privacy rights or ask about this policy, contact leyinzylin@gmail.com.

Accounts, membership, and security

The app can create an account without registration. To establish sessions, prevent abuse, and determine eligibility for the one-time welcome allowance, we process account and installation identifiers, Apple-signed AppTransaction evidence, and necessary network information.

When you register or sign in, we process your email address and account identifier. For paid features, we process Apple-verified product and transaction details, subscription status, and entitlement records. These data support authentication, membership and usage billing, account security, and support. Email addresses and verification codes are delivered through Resend when you sign in by email.

Local content, iCloud, and voice input

Conversations, notes, folder names and hierarchy, and reminders are stored on your device by default. Every account can enable iCloud sync. When you enable it, this content is stored in your private iCloud database and processed by Apple. Turning off sync stops future synchronization. Deleting your saidosi account does not automatically delete this content from your device or iCloud.

Starting with 1.5.0, when you enable microphone dictation or continuous conversation and grant system microphone permission, the app sends microphone audio directly from your device to Tencent Cloud realtime speech recognition for transcription within China. The audio does not pass through saidosi servers, and microphone dictation does not save recordings. According to Tencent Cloud, incoming audio is used only for the current recognition and is not retained. Speech recognition happens before you send an AI request. Once you send the recognized text, it is handled in the same way as typed text. Earlier versions use Apple speech recognition services for microphone dictation.

To provide realtime speech recognition, record usage, and calculate costs, saidosi servers retain speech grant and session identifiers, audio duration, recognition status, and estimated costs per account. These records contain no audio or transcription content. Historical usage and cost records remain subject to the retention rules for billing and security records after account deletion. Imported audio attachments and Apple Watch recordings are transcribed using Apple's on-device speech models.

AI requests

Before your first send from the input field, saidosi explains what data will be sent, why it will be processed, and the third-party LLM providers receiving it: DeepSeek, Zhipu AI (GLM), and Tencent Cloud TokenHub. It then asks for your explicit permission. AI data processing is off until you allow it on the current installation, and your current input is sent only after you allow it. You can review or change your choice in Settings → Service Agreements → AI Data Processing Notice. Local features remain available without permission.

With permission, we process the text, audio transcriptions, and images you enter or choose to send, conversation content and related notes needed to understand the request, folder names and paths, and the tool operations, parameters, and results needed to complete it. This content uses one AI permission; existing permission needs no additional image confirmation. Related notes provide context for replies. Do not submit sensitive information that is unnecessary for the task.

Text and other request content are forwarded through saidosi servers; images use the version-specific temporary upload paths described below. Model services are provided by DeepSeek, Zhipu AI (GLM), or Tencent Cloud TokenHub and process requests within China. We do not send your account identifier, email address, phone number, or device identifier to the recipient. Personal information you include in text or images is sent with that content. AI requests are not used for advertising or cross-app tracking. These rules are collected in the AI Data Processing Notice.

When a task needs web search, search queries are sent through saidosi servers to DeepSeek. To read a public webpage, the app connects directly from your device to the destination website. The website may receive network information such as your IP address and handle it under its own privacy rules. Retrieved page content may be used as tool results for the current AI task.

saidosi servers do not retain AI request bodies or use requests or replies to train models. To let you continue receiving results after a disconnection or while the app is in the background, servers may temporarily cache reply streams per account, including reply text and tool-call data. A cache is available for resuming for 24 hours after creation. After expiry, it can no longer be resumed and a background task removes it; service failures may delay removal. Deleting an account also deletes its response caches. Billing and security records do not contain bodies.

DeepSeek, Zhipu AI (GLM), and Tencent Cloud TokenHub handle content under their own terms and privacy policies. They may retain necessary records and use content to generate results, manage security and compliance, or improve services or models. Their policies govern the specific limits on retention and human review.

You can withdraw permission at any time in Settings → Service Agreements → AI Data Processing Notice. Withdrawal stops new image uploads. Your next send from the input field will ask for permission again. Onboarding, retries, tool or startup continuations, and system entry points do not show an additional permission prompt. Images from history or tool results do not trigger another prompt. Withdrawal does not stop work that has already started. Content already sent and processed by a recipient cannot be recalled. Account deletion, local content deletion, iCloud controls, and subscription cancellation are separate actions.

Fatal crash reports

If the App Store version has a fatal crash, it automatically sends a crash stack, app and system versions, device model, IDFV, network status, region, and related diagnostic information to Tencent Bugly. After sign-in, reports also include the account identifier and sign-in email to help diagnose reliability issues. Crash reports exclude conversations, notes, AI inputs and outputs, and feedback attachments. They are not used for advertising or cross-app tracking.

Feedback and diagnostics

All users can submit feedback in the app, including unregistered users and users without a service account. We process feedback text, optional contact details, app and device environment, and diagnostic attachments you choose to include. After the service update for version 1.4, new feedback is delivered through Resend to our Gmail support inbox; beforehand, it is received by a private support system hosted by GitHub. Feedback is used only to respond, investigate issues, and verify fixes.

Runtime diagnostics are stored locally per installation and current account, are not backed up or continuously uploaded, and are retained for up to approximately 30 days. They include runtime state, error categories, counts, durations, and opaque identifiers. Conversation, note, and AI bodies are not written to diagnostic logs. Diagnostics still within their retention period are uploaded only when you keep the attachment and send feedback. You can remove it before sending. Temporary attachments are deleted after removal, sending, or abandonment; source diagnostics keep their normal retention period. Signing out or deleting an account removes that account's local diagnostics.

After the service update, servers extract only log-level counts and up to 100 request identifiers from diagnostic attachments for the email summary, without forwarding log bodies. Raw attachments are removed after the request and are not sent to email or GitHub. Feedback emails, contact details, and handling records are retained as long as needed; unresolved feedback, or feedback that will not be acted on, may be retained long term. Historical GitHub feedback and attachments keep their existing lifecycle: once a fix is released and confirmed, we delete the corresponding diagnostic attachments but may retain necessary feedback and handling records. There is currently no separate deletion process for unresolved historical feedback and attachments.

Third-party processors

We use these services only as needed for the features above and apply reasonable measures such as encrypted transport and access controls. We confirm that third parties receiving user data will provide the same or equivalent protection required by this policy and applicable law under their agreements and privacy rules, and process it only for the disclosed purposes and as permitted by law. If processors or purposes change materially, we will update this policy and obtain renewed consent when legally required.

Temporary image storage

AI data processing permission includes image uploads and temporary storage, with no separate confirmation. With permission and an available account, starting with 1.5.0, images may be uploaded after selection, before you send them, directly from your device to a private temporary Tencent Cloud COS bucket in Nanjing, China. The model provider reads images through temporary authorized links. Image bodies are not forwarded through saidosi application servers, which do not store image reference records. Selecting an image does not request permission. Without permission, image drafts stay on your device until you send and allow AI data processing.

COS image references used by 1.5.0 and later can be reused for seven days after upload authorization. Cloud storage asynchronously removes temporary copies after seven days from upload; actual deletion may be delayed.

Supported versions 1.3.x and 1.4.x use a different path: images are sent from your device to saidosi servers and forwarded to DeepSeek Files temporary storage, requesting a file lifetime of about one hour. The returned temporary reference can be reused until it expires, after which it cannot be used for new model requests. saidosi servers do not persist image bodies or reference records. Actual removal of provider files follows its service rules and may be delayed.

Removing a draft does not immediately delete uploaded copies in either path. Local originals are unaffected by expiry and can be uploaded again when needed. Withdrawing permission stops new uploads. Processed content cannot be recalled; removing temporary copies does not also delete provider caches, security records, or other processed copies.

Retention, deletion, and choices

While your account exists, we retain the account details needed to provide services. You can delete your account in Settings. Deletion clears the sign-in email and account identifier and revokes sessions. Transaction, settlement, fraud-prevention, and necessary audit records may be retained for legal and security needs, separately from deleted direct identity information. Deleting an account does not cancel App Store subscriptions or delete conversations, notes, folders, or reminders from your device or iCloud. Manage that content first in the app or Apple's iCloud management tools.

You can choose not to enable iCloud or voice input, omit feedback diagnostics, stop using AI, or withdraw AI permission in Settings → Service Agreements → AI Data Processing Notice. Withdrawal does not delete your account or local content, disable iCloud, or cancel subscriptions. Account deletion, local content deletion, iCloud controls, subscription cancellation, and AI data processing permission are independent choices. You can also email leyinzylin@gmail.com to request access, a copy, correction, or restriction of processing of other personal information, or account closure. We may need to verify your identity and will explain legal reasons why we cannot fulfill a request in full.

Children

saidosi is not intended for children under 14. If we discover that a child's personal information was processed without valid guardian consent, we will delete it or take other necessary measures as required by law.

Policy updates

Updates will appear on this page. If processing purposes, methods, or data categories change materially, we will notify you before the changes take effect through appropriate means, such as an in-app notice, and obtain renewed consent when required.