What is Navōn? The kosher AI platform, explained

Navōn, also written Navon, is a kosher AI platform for schools, families and frum workplaces. It lives at navon.org. The name is the Hebrew נָבוֹן, which means wise or discerning, and the second meaning is the one we had in mind: a tool that can tell the difference between what belongs in front of a particular person and what does not.

This page sets out what Navōn is, who it is for and how it works, in enough detail that you can check it rather than take it on trust.

A note on the name

The line over the o is there to show the pronunciation, with the stress on the second syllable. Written without it, as Navon, it is the same name and the same platform.

There are other companies and products called Navon, in a range of fields. Navōn is not connected to any of them. If you were looking for one of those, this is not it. If you were looking for the AI platform at navon.org, it is.

What Navōn is

Navōn is a general-purpose AI assistant. You can ask it questions, have it help with writing, translation, maths and code, summarise a document, generate and edit images, work with video, and have answers read aloud. In that respect it resembles the tools you have already heard of, and it is not built on a weaker model to make filtering easier. On Auto, it reads each request and routes it to one of several leading models from Anthropic, OpenAI and Google, and shows which one answered.

The difference is what surrounds the model. Every request passes through filtering on Navōn's servers before it reaches a model, and every answer is checked before it is accepted. The standard it is checked against is not one global policy written for everyone at once. A core set of rules applies to every account, and the school or parent adds to it.

Who it is for

• Schools. A menahel or technology coordinator sets one filter policy for the whole school, and staff can see what was asked and what was declined. Classes and assignments sit in the same platform, so talmidim do their schoolwork under the school's rules wherever they sign in.
• Families. A parent sets up an account for each child, blocks the topics the family does not want discussed, and can set allowed hours and a daily limit.
• Students, who need help with homework and a tool their parents and school are comfortable with.
• Frum workplaces and individuals who want AI for writing, research and parnassah, without the material a general-purpose assistant will produce for anyone who asks.

How it works

Filtering on the server, on both sides

The checks run on Navōn's servers, not in the browser and not as an extension on a particular device. That makes the filtering a property of the account rather than of the machine. A talmid who signs in on a phone gets the same rules as on the school computer, and there is nothing on the device to switch off.

Both sides of the exchange are checked. The prompt is checked before it reaches a model, including the text of any document attached to it and any picture uploaded with it. The answer is scanned when it is complete, and an answer that fails is replaced on screen with a notice rather than left in place.

Standards the school or parent sets

Some categories are locked for every account and cannot be switched off by anyone, including an administrator. Above that floor, a guardian or school can block topics of their own, in their own words, and set categories to blocked, guided or allowed.

A household can also say which tradition it follows, such as Litvish, Chassidish, Sephardi or Modern Orthodox. Where a question in halacha or hashkafa is answered differently by different communities, the answer gives that household's practice first and says plainly that others hold otherwise.

Images, video and voice under the same rules

The media tools are not a side door. Image and video prompts pass through the same check as chat, and a generated picture is scanned before it is stored, so one that fails is never saved. People in generated pictures are dressed modestly, and where a household has set its tradition, in the manner of that community.

You can dictate a question in English or Hebrew and have an answer read aloud. What is spoken becomes the same text, and goes through the same checks, as what is typed.

Guardian controls

A parent can see what was asked and what was declined. Allowed hours and a daily limit are checked on the server, so they hold on any device. The homework and study helpers can be switched off. When a request is declined, the reply appears in the conversation and suggests something else to ask, rather than a warning box that ends the exchange.

If a child writes something suggesting they are being hurt, or are thinking of hurting themselves, the ordinary rules are set aside so that the child gets a warm reply pointing them to a trusted adult, and the guardian is emailed.

Fail closed

Every filter depends on a check that will sometimes be unavailable. When Navōn's safety check cannot run, the request is refused with a message saying so, rather than answered unchecked. During an outage that means nobody can work for a while. For a tool children use, that is the right trade, and it is the one the system makes.

What it does not claim

No filter is perfect, and Navōn does not claim to be. Content moderation is judgement at a boundary, done at scale, and every system misses something sometimes. When something does get through, it can be reported, and a record exists for the guardian or school to look at.

Navōn is not a source of psak. It can explain a passage, translate, help prepare a dvar Torah and set out what different sources say. It cannot tell you what to do, and a practical sheilah belongs with your rav. There is no hechsher for software, and Navōn does not carry one.

It also covers only itself. It cannot reach a different tool open in another tab, and it does not replace a parent or teacher who knows what a child is using.

How to start

Navōn is opening in stages, with plans for individuals, families and schools. Every plan includes the same server-side filtering on every prompt and every response. We would rather you tested that filtering against your own standards than took this page's word for it.