Add jevos

[feder-cr]

Project: https://github.com/feder-cr/jev

What it uses Jev for: It doesn't call Jev; it's an open, MIT-licensed replacement for Jev's yes/no (Noul) decisions, run on a laptop CPU.

Placed in Open Models & Reproductions rather than "Running Without Jev": your own README says that section is for tools with a configurable endpoint, and that model replacements like this belong in Open Models & Reproductions instead, so the "actually calls Jev" checkbox below doesn't apply here (same as other entries in that section, e.g. jeff, kev, OpenJev).

• The project actually calls Jev / TypeSafe System One, not just mentions it. (N/A: it's a model replacement, see above)
• It has a README that explains what it does and how to run it.
• It's in the most relevant section, using the format - [Name](link) - Description.
• Any speed, cost, or accuracy numbers are the author's own and marked as author-reported.
• I read the contribution guidelines.

Disclosure: I'm one of the project's authors (Federico Elia).

🤖 Generated with Claude Code

[MrJev]

Landed in #52 — thank you, and thank you for arguing the section from our own README and saying up front that it makes no Jev call. That made this a ten-minute job instead of an hour's.

What we checked before listing, in the code rather than from the description: jev serve speaks POST /v1/systemone and GET /v1/models; the CLI's own help says the binary model "answers yes/no (noul) questions only, as 1 or 0; other types are refused", so the scope is enforced rather than only documented; the quickstart fetches a llama.cpp runtime and serves a GGUF with --device cpu, and nothing needs the network after that. 886 stars, MIT, pushed today.

One edit to the line, and the reason: your description said "a 1B model cut to 17 layers", and we could not find that stated anywhere in the repository — not the README, not the release notes, not the loader. Our rule is that a number in an entry has to be sourceable, so the published description leaves the architecture out and keeps everything else you wrote:

• jevos - Serves /v1/systemone and /v1/models for yes/no questions only — other types are refused — from a GGUF model through llama.cpp on CPU, offline once the file is downloaded.

If the layer count is written down somewhere we missed, link it and we will put it back. It would be a good thing to have in the README either way: "which base, cut how" is the first question a reader of this section asks.

It will appear on mrjev.com/projects after the next data sync. The hands-on review is a separate queue and runs behind listing — the page will say "no review yet" until then, which is normal rather than a judgement.