# תוכנה פלוס

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [yehonatan shoham](https://cerebralvalley.ai/u/yonicc)
- **GitHub:** https://github.com/yonicc/lashon-hara-guard
- **Demo video:** https://youtu.be/IWiK9mE4CHI
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/573

Lashon Hara Guard — A browser companion that helps people pause before harmful gossip.

Lashon Hara Guard helps people reconsider potentially harmful messages before sending them in Gmail, WhatsApp Web and Google Chat. It raises a gentle warning, asks targeted clarifying questions, and can suggest questions to bring to a qualified rabbi. Incoming messages can be manually checked before they are revealed. English and Hebrew are supported. It is a reminder and decision-support tool, never a halachic ruling.

Everyday digital conversations make it easy to spread damaging gossip in seconds. Recognizing the risk depends on context: what was actually said, who the audience is, what the sender knows and whether essential facts are missing. A useful assistant must ask rather than guess, while leaving the final decision with the person.

A lightweight gate classifier routes potentially relevant text to a Meta Muse Spark extractor. Proposed facts carry verbatim quotations and pass validation before a deterministic engine consults source-linked draft rule cards. Missing facts remain unknown. The system can ask up to five catalog questions or offer a referral question list. The extension keeps locally scoped context and supports cloud, hybrid and local configurations; cloud checks transmit message content for processing.

Try it: https://lhg.ysho.uk (Chrome and Firefox, v0.17.4).

The rule cards are drafts awaiting qualified review. The tool never declares a message permitted, cannot establish external facts, and does not replace expert judgment. The demo does not send an email. Cloud processing is disclosed in PRIVACY.md.

JUDGE TEST ACCESS (reviewer credential; please do not republish):
Download the Chrome or Firefox extension from https://lhg.ysho.uk/downloads/. Configure the extension with this judge access token: lap_hgqme5CpdpgDC-_zdxPTSdvz6Y36cscKLQSlwnitPSU
The judge account is billing-exempt, with a safety cap of 500 checks per day. Open Gmail, WhatsApp Web or Google Chat; for a safe test, compose a fictional draft with no recipient and invoke the outgoing check (Ctrl+Enter in Gmail). See the real warning in the demo. The tool is a reminder, not a halachic ruling.

## More from Meta Global AI Developer Hackathon

- [LOCAISH](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/570)
- [The Hidden Layer](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/571)
- [Software Fanatics](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/572)
- [FlyPig AI](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/574)
- [Makabeez](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/575)
- [JL](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/576)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/573. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt