How Heimish AI Works

When you type a message, Heimish AI sends your text — along with your recent conversation history and any files you've attached — to a large language model. The model is a mathematical system trained on vast amounts of text: it predicts what words should come next based on patterns it has learned.

Before the model ever sees your message, it passes through a Torah-based moderation layer. This layer checks your input for inappropriate requests — requests that violate Torah values — and blocks them before they reach the model. The model's response is checked again on the way back. If the response contains something inappropriate, it's blocked and you see a message explaining why.

The assistant remembers your conversation so it can give coherent, contextual answers. You can start a new conversation at any time, and old conversations are saved to your private library so you can return to them.

Heimish AI can learn to write in your voice. When you upload writing samples — articles, emails, or past conversations — to your private library, the assistant uses them as a reference for tone, vocabulary, and style. Your next response will sound more like you, not like a generic AI.

Your writing samples are private. Only you can see them, and they're never shared with other users. You can delete them at any time, and you can turn style learning off completely in your settings.

When you switch to image mode and describe what you want, Heimish AI sends your prompt to an image generation model. The model creates a new image from scratch based on your description — it doesn't search for or modify existing photos.

Just like text, image prompts pass through the moderation layer. Requests for inappropriate images are blocked. Generated images are saved to your private gallery, and you can choose to share them in the public gallery for the community to see.

This is the core of what makes Heimish AI different from an unfiltered AI. Every request — text and image — is checked against a set of Torah-based guidelines before it's processed, and every response is checked again before it's shown to you.

The moderation layer blocks: inappropriate or explicit content, requests for halachic rulings (the AI cannot pasken), lashon hara about Jews or private individuals, and content that mocks or disrespects Torah concepts. When something is blocked, you see a clear message explaining why.

Important: The AI never provides halachic rulings. If you ask a halacha l'maaseh question, you'll receive a reminder that the AI cannot pasken and a suggestion to consult your local rav. This is by design — the AI is a study and productivity tool, not a halachic authority.

Your conversations, writing samples, and generated images are tied to your account and private to you. Other users cannot see your chats or your private gallery. Administrators can see conversations for moderation purposes — to review blocked content or handle abuse reports — but your data is not shared or sold.

You can delete any conversation, writing sample, or image at any time. Deleting is permanent. You can also delete your entire account from your settings, which removes all your data.

Heimish AI works in English, Hebrew, and Yiddish. The interface switches to match your chosen language — including right-to-left layout for Hebrew. The assistant responds in the language you write in, so you can switch languages mid-conversation and it will follow.

Heimish AI offers a free model (automatic) and premium models (Claude, Gemini Pro). Free questions and images are available every day. Premium models and additional usage cost credits, which you can buy or earn with an unlimited subscription.

See the pricing page for the full breakdown of credits and plans.