One question, two answers, zero cloud

What it does

jev — the description calls it jevos — is a self-hosted alternative to Jev for yes/no decisions. You feed it text and it returns one of two answers. The repo frames this as binary text classification, done by a small local language model rather than a hosted API.

The interesting bit

The stack is the story: a quantized MiniCPM model in GGUF form, run through llama.cpp on CPU, served via FastAPI. A general-purpose LLM pressed into duty as a two-label classifier, with the entire pipeline staying on your machine.

Key highlights

• Open-source alternative to Jev for yes/no decisions, per the repo’s own description
• Runs fully offline on a laptop — CPU inference, no outbound calls, no API keys
• Small LLM under the hood: MiniCPM as a GGUF file via llama.cpp, per the repo’s topic list
• Exposed as a FastAPI service and positioned as edge/on-device AI
• Self-hosted by design

Caveats

• The README linked from the repo documents an entirely different project (Resume_Builder_AIHawk, a resume-tailoring tool), so actual usage, model specifics, and any benchmarks are unclear from the provided sources
• What “Jev” itself is — the thing being replaced — is never explained
• GitHub lists the language as C++ while the topics point at Python and FastAPI; the real split is anyone’s guess

Verdict

If you want a private, offline yes/no classifier — or a working template for serving small GGUF models behind a tiny API — this is worth a look. If you need documented behavior or benchmarks, the sources here don’t provide them, and the mismatched README won’t help.