KosherAgent.

Legal · AI

AI Use Policy

Last updated: 2026-05-10

KosherAgent is a voice assistant powered by artificial intelligence. This page explains, in plain terms, what AI is doing in the call, what it’s good at, what it’s not good at, and how we handle the data you give it.

1. AI is in the loop

When you speak to KosherAgent, AI models are part of the call path. Specifically:

  • A speech-recognition model converts your spoken words to text.
  • A reasoning model decides how to respond and which tools to call (look up zmanim, place an order, set a reminder, send a text, etc.).
  • A speech-synthesis model speaks the response back.

The result is a voice conversation. There is no human listening to a typical call in real time, though our security and quality teams may review aggregated, de-identified samples to debug problems and improve the service.

2. What AI is good at

  • Understanding everyday spoken English (and basic Hebrew / Yiddish phrases).
  • Following multi-step instructions in a conversation.
  • Looking up factual information from connected tools (zmanim, weather, calendar).
  • Drafting short messages and translations.
  • Confirming details before taking irreversible actions.

3. What AI is not good at

You should not rely on KosherAgent — or any AI service — for:

  • Medical decisions. AI cannot diagnose, treat, or prescribe. Talk to a doctor.
  • Legal advice. Talk to a lawyer.
  • Financial advice. Talk to an accountant or financial advisor.
  • Halakhic rulings. Ask a rav.
  • Safety-of-life situations. In an emergency, dial 911. KosherAgent is not designed to handle 911-style calls.
  • Recent events. AI models have a knowledge cutoff. We connect tools to give you current information where we can, but anything time-sensitive should be verified.
  • Highly technical or specialized work. AI can summarize but it can be wrong about specifics.

We confirm before any irreversible action (placing an order, sending a message, canceling an appointment) but you are the final approver. Listen to the confirmation.

4. AI errors are possible

AI models make mistakes. They can:

  • Misunderstand what you said (especially in a noisy room).
  • Confidently state something inaccurate (“hallucinate”).
  • Miss context from earlier in the conversation.
  • Respond in unexpected ways to ambiguous requests.

If KosherAgent gets something wrong, please tell us. The fastest way is during the call (“That’s wrong, please correct it”). You can also email support@thekosheragent.com with the call time and a brief description.

5. Your inputs & ownership

  • You own what you tell KosherAgent. Your spoken requests, the calendar entries you create, the reminders you set — all yours.
  • You grant us a license to operate. To deliver the Service we need to process what you tell us. That license is limited to operating, maintaining, and improving the Service for you.
  • You own the outputs we generate for you (drafted texts, summaries, translations) and may use them for any lawful purpose.

6. We do not train AI models on your conversations

This is important enough to be its own bullet:

We do not use your voice, your transcripts, or your derived data to train AI models — ours or any third party’s.

We may use aggregated, de-identified statistics (e.g., “calls about zmanim averaged 14 seconds last month”) for product improvement. These statistics cannot be linked back to you.

If we ever change this policy, you will be notified by email and in-call announcement before any change takes effect. You will have the option to opt out and have your data deleted.

7. You may not train competing models on our outputs

You may use KosherAgent’s outputs for your own purposes. You may not use them to train, fine-tune, or evaluate a competing AI service. This protects the integrity of the model layers we build on.

8. Limitations of voice biometrics

KosherAgent can recognize enrolled household members by voice. Voice biometrics are not perfect:

  • A bad cold or a phone with poor mic quality can fail to match.
  • Voice biometrics alone are not enough for sensitive actions. We require an additional factor (a spoken PIN or callback verification) before changing account settings, transferring money, or other high-risk operations.
  • We do not use your voiceprint outside the KosherAgent service. We do not sell or share it.

9. Disclosed use of AI

If a partner or business uses KosherAgent’s voice technology to power their own customer-facing service, that partner must clearly disclose to their callers that they are speaking with an AI. We make this a contractual requirement of using our APIs.

10. Reporting harmful or wrong outputs

If KosherAgent produces output that is harmful, biased, hateful, illegal, or just deeply wrong, please report it to safety@thekosheragent.com. Include the approximate call time and what was said. We review every report and adjust our safety filters as needed.

11. Changes to this policy

If we change how we use AI in a way that materially affects your privacy or safety, we will update this page and notify you by email or in-call announcement.