AI Policy
1. Which providers, and how they are called
Readings are written by DeepSeek (deepseek-chat) as the primary model and by OpenAI (gpt-4o) as an automatic fallback when the primary fails or is unavailable. Both are called directly from our servers over their public APIs; no cloud intermediary sits in between.
| Provider | Role | Model | Where |
|---|---|---|---|
| DeepSeek (Hangzhou DeepSeek Artificial Intelligence Co., Ltd. (China)) | Primary | deepseek-chat | China |
| OpenAI (OpenAI, L.L.C. (United States)) | Fallback | gpt-4o | United States |
2. Exactly what is sent, and what is withheld
Sent with each reading request:
- the name and gender you entered for the person being read, exactly as typed
- that person's birth date, time and place, and the computed chart (planetary positions, houses, nakshatras, dashas)
- the reading language and the report sections you requested
- any question you type for the reading or the report chat
Not sent:
- your email address and account identifiers
- client contact details (email, phone), notes and family details
- payment details
This is not anonymous data. A date, time and place of birth together identify a person, and the name you typed for that person travels with the chart. Enter only what you are entitled to share, and use an initial instead of a full name if you prefer.
3. Training and retention at the provider
We rely on each provider's API data terms, under which content sent through the API is not used to train their models. Providers may retain requests briefly for abuse monitoring under those terms. We have no contract that goes beyond the standard API terms; if that changes, this page will say so. Our own record of each request keeps only metering data (model, token counts, cost, latency), not the content.
4. Human review
A reading is a draft produced by a language model. It is not reviewed by an astrologer before you see it. You are responsible for reading it critically before you rely on it or pass it to a client.
5. Accuracy and bias
Models can state things confidently that are wrong, can reflect biases in the texts they were trained on, and can misread an unusual chart. The chart calculations themselves are deterministic and reproducible; the interpretation is not.
6. Provider changes
If we add, remove or reorder a provider, or a provider moves region, we publish a new version of this policy and of the Third-Party Services page. Questions: contact@ipandith.com.