Jev4Mellea
Unofficial Python adapter that brings Jev verification, classification and scoring into the Mellea framework as Requirements.
Overview
Mellea x Jev is a small synchronous Python adapter that brings TypeSafe Jev's semantic checks into Mellea, a framework for structured generative programs. Jev evaluates text; Mellea continues to manage generation and repair. You can verify requirements with a Noul and configurable accept and reject thresholds, classify text into caller-defined categories with a Choice, score text on an ordered scale, ask several questions in one request and inspect token usage, and attach checks to Mellea Requirements so they participate in sampling.
The README's example verifies whether a candidate answer gives a museum's opening time from a reference passage and returns pass, fail or uncertain with the probability of yes. Requests can go through a proxy with a TypeSafe-compatible API root. A demo runs without a key or network using a mocked response, it has CI, and it is Apache-2.0 licensed.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Does this candidate meet the requirement (e.g. gives the museum's opening time from the reference)? | — |
| Choice | Which caller-defined category does this text belong to? | — |
| Score | Where does this text fall on an ordered scale? | — |
How it works
- Mellea generates candidate text
- Jev verifier evaluates it against a requirement
- Thresholds yield pass, fail or uncertain
- Mellea uses the result for sampling or repair
Strengths
- Plugs into Mellea Requirements
- Three-way outcome with uncertain band
- Offline mocked demo
Caveats
- Unofficial and synchronous only
- Live requests incur charges
Getting started
git clone, make install, export TYPESAFE_API_KEY, or `make demo` to run offline.At a glance
- Type
- Project
- Status
- beta
- Language
- Python
- Runtime
- Library (Mellea adapter)
- Has code
- yes
- Has tests
- yes
- License
- Apache-2.0
Use cases and integrations
How curated lists describe it
This is an unofficial, synchronous adapter. Jev evaluates text; it does not generate or repair it.
logicrw/awesome-jev-projectsJev adapter to Mellea.
hellogumbo/awesome-jev
Frequently asked questions
Does Jev generate or fix text here?
No. Jev only evaluates; Mellea handles generation and repair.
What outcomes does verification return?
pass, fail or uncertain, with the probability of yes.
Can I try it without a key?
Yes, `make demo` uses a mocked response.
Related projects
Companion code comparing Google ADK and Microsoft Agent Framework for structured video timelines, with Jev as a shared per-segment QC gate.
LibraryprototypePythonUiPath bring-your-own-guardrail connector that runs AI Trust Layer llm_as_judge checks as Jev noul questions.
ProjectprototypeJavaScriptAI SDK middleware and OpenAI-compatible proxy that runs seven Jev checks on every LLM response, gated by confidence.
ProxyprototypeTypeScriptPilot that puts Jev into LangGraph routing and Deep Agents guardrails and claim verification, with executed notebooks and caveats.
ProjectprototypePythonTerminal Q&A app where Jev sets response policy and reviews drafts while OpenAI only writes, orchestrated with LangGraph.
AppprototypePythonPublished Jev client and claim check from the Fake/Real evidence-review site, where Jev filters sources and judges each claim.
ProjectproductionTypeScript