Privacy Policy
Effective date: [launch date] · Applies to Chengzhang (international edition)
Chengzhang ("we") turns what you say into written articles. This policy explains what we collect, why, who processes it, how long we keep it, and what you can do. We only collect what Chengzhang needs to work.
Operator: 上海子微星科技有限公司 Contact: contact@xingziai.com
What we collect
| Data | Why |
|---|---|
| Your email address, or the email Google shares when you sign in with Google | Sign-in, account recovery, essential service notices |
| Voice recordings | Transcription. By default a recording is deleted as soon as it has been transcribed; only the text is kept |
| Material you add: transcripts, pasted text, links, images, files (PDF, Word, etc.) | Organizing material and writing your article |
| Articles, version history, translations, annotations, your ratings | Editing, saving and publishing your articles |
| Your style notes, vocabulary and the past articles you provide | Making drafts sound like you |
| A record of each AI call: which step, how much it used, how long it took, whether it failed | Usage limits, cost accounting, troubleshooting |
| IP address and browser information | Sign-in security and abuse prevention (rate limiting) |
We do not sell your data, show ads, or use third-party advertising or tracking tools. We only use the cookies needed to keep you signed in.
Who processes your data
Chengzhang does not train models. To provide the service, the following providers process some data on our instructions:
| Provider | What | Why |
|---|---|---|
| Anthropic (Claude) | Material, style references, articles, your edit instructions; web searches during fact-checking | Organizing, writing, fact-checking, editing, translating |
| OpenRouter and the speech model provider it routes to (currently Alibaba Cloud Qwen) | Recordings | Speech-to-text |
| LangChain (LangSmith, Asia-Pacific region) | Steps, models, usage and timing of each AI call; with "Help improve Chengzhang" on, also the input and output text | Troubleshooting and improving writing quality |
| WorkOS | Email, IP address and browser information at sign-in | Sending sign-in codes, Google sign-in |
| Dodo Payments, Apple (App Store) | Top-up amounts and billing details; card details are collected by them directly and never reach Chengzhang | Payments, invoicing and taxes |
| Google (only if you choose to sign in with it) | As part of their sign-in flow | Verifying your identity |
| Railway (Singapore) | Storage and hosting of all data | Servers and database |
Under Anthropic's commercial terms, content submitted through its API is not used to train its models. Each provider handles data under its own privacy policy and data processing terms. Some providers are located in the United States or other countries, so your data may be transferred internationally.
When you add a link as material, our server fetches the page. During fact-checking Claude searches the web, and search queries may include keywords from your article.
Public image links
Images you add to material or articles get a long, unguessable public link (/i/…) that opens without signing in, so that images come along when you paste an article into WeChat, Substack and other platforms. Anyone with the link can see the image, so do not put anything private in images.
Help improve Chengzhang (off by default)
You can turn this on in Settings → Help improve Chengzhang. When it is on, articles you rate "good" or "bad" are used to improve our writing quality:
- What is used: the article's material (transcribed voice notes, links, images, files), the AI first draft, your final version, your rating and comment, and your style notes and vocabulary. Records of the writing process (each step's model input and output) include this text too.
- What is removed first: email addresses, phone numbers, ID and bank card numbers, and your byline and email username. Other people's names mentioned in the text are not removed automatically.
- Who sees it and why: only the Chengzhang developers, to test and compare new writing approaches. We do not publish it, sell it, or provide it to third parties for model training.
- Where it is stored: samples and process records are stored in LangSmith (Asia-Pacific region), the analytics service we use. Data from the mainland China version stays in mainland China.
- How to withdraw: turn it off at any time. Your articles are removed from our samples immediately, and future process records contain no text; existing records expire automatically (within about 400 days). The same happens when you delete your account.
Whether or not you turn it on, we keep anonymous usage data: which features you use and how often, how much of the first draft you changed, whether AI annotations were kept, which cover style and publishing platform you used, and payment events such as top-ups or running out of credit. These records contain only event names and numbers—no text, titles, or URLs—and are not linked to your email (a one-way anonymous ID tells users apart). They are stored in our own database in your region and are not sent to third parties. They are deleted when you delete your account.
When the setting is off, writing-process records keep only the steps, models, usage, and timing—not your material or text.
Feedback: text and screenshots you send via Settings → Feedback are stored with your account email and used only to reply to you and improve the product.
How long we keep data
- Recordings: deleted as soon as transcription succeeds (kept until retries finish if it fails).
- Account, articles, material and style records: kept until you delete your account. Deleted articles and material are permanently removed after 30 days.
- Up to the 60 most recent versions are kept per article.
- AI call records (LangSmith): usually 14 days; samples and records used for improvement up to about 400 days.
- Sign-in sessions expire after 60 days. IP counters used for rate limiting are cleared within a day.
- Top-up and usage charge records: after you delete your account, we remove information that identifies you and keep them for 7 years for accounting and tax purposes.
- Database backups: kept for up to 30 days, then deleted automatically.
Your rights
- Export: download all your data (articles, translations, original material files and full records) under Settings → Account.
- Correct and delete: edit or delete articles, material and style records directly.
- Delete your account: under Settings → Account. Your account and all content are removed from our database immediately, along with your identity at our sign-in provider; backups and LangSmith records expire on the schedule above.
- If the law where you live (for example the GDPR or the CCPA) gives you further rights, such as access, restriction, objection, or complaining to a regulator, email us and we will reply within 30 days.
Security
All data is encrypted in transit (HTTPS). Passwords, if any, are stored only as salted hashes, and sign-in tokens are stored only as hashes. Only people who need to maintain the service can access production data.
Children
Chengzhang is not intended for anyone under 16. If you learn that someone under 16 is using it, tell us and we will delete the account.
Changes
If this policy changes materially, we will tell you in the app or by email in advance.
Users in mainland China
This is the international edition. Data is stored in Singapore and processed by the providers listed above outside mainland China. A separate edition that stores data within mainland China is planned.