jevify (lucioamor)
Skill and MCP service that audits an app for AI calls that are really decisions and shadow-tests migrating them to Jev.
Overview
jevify (lucioamor) addresses the practical question behind claims of savings from moving decisions off general-purpose LLMs: which calls in your app, which lines, what changes, and how you would see the gain. It works in three inspectable steps. An audit points at real call sites, shows how each response is consumed, and separates bounded decisions (a queue, a score, a yes/no) from generation that should stay generation; a clean audit is a valid outcome. Each candidate gets a before-and-after sketch naming the primitive, the TypeSafe pattern, the closest official cookbook and what the app would need to adapt. Then a migrate command runs the Jev path in shadow beside the current one, logging cost per decision, p50 and p95 latency, agreement, labelled disagreements and fallback rate, so the gain becomes a number from your data or the migration stops.
The skill is the canonical source; a hosted MCP service runs the same audit and keeps reports under your account; an open-source local MCP server runs the engine offline; and a Lovable skill brings it into Lovable projects. Every effect is treated as a hypothesis until shadow data says otherwise.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which queue should this ticket be routed to? (example migrated decision) | sales · support · other |
How it works
- Audit the codebase for decision-shaped AI calls
- Sketch the Jev replacement per candidate
- Run the new path in shadow and log both
- Adopt only if measured cost, latency and agreement justify it
Strengths
- Evidence-driven migration on your own traffic
- Separates decisions from generation
- Hosted and local MCP options
Caveats
- Shadow mode doubles calls during testing
- Hosted service keeps reports under an account
At a glance
- Type
- MCP server
- Status
- beta
- Language
- TypeScript
- Runtime
- Agent skill + MCP server
- Has code
- yes
Use cases and integrations
How curated lists describe it
Jevify: find the AI calls in your app that are really decisions, then migrate them to JEV in shadow mode. Skills for Lovable and Claude Code, backed by the jevify MCP.
hellogumbo/awesome-jev
Frequently asked questions
What does the audit find?
Call sites where an AI response is only used as a bounded decision, separated from real generation.
How is the gain proven?
By running the Jev path in shadow and logging cost, latency, agreement and fallbacks.
Can it run offline?
Yes, via the open-source local MCP server.
Related projects
Local code search for coding agents: ripgrep/BM25 shortlist, then Jev judges each chunk's relevance, served over MCP.
MCP serverbetaRust- jevgrep (dzhng)★ 2,671
Code search CLI and agent skill (jg) where Jev judges relevance across folders, files and declarations to return excerpts.
Agent skillbetaTypeScript Self-hosted Jev MCP server for Claude Code and Codex with confidence gating, cost guardrails, caching and decision logs.
MCP serverbetaPythonKey-locked Netlify proxy that serves Jev through Netlify AI Gateway credits, plus an MCP server for Claude Code, Codex and Claude Desktop.
MCP serverbetaPythonClaude Code plugin that finds fuzzy decision points in an app and ranks where a Jev question could replace them, with precedents.
ProjectbetaPythonRust CLI for Jev that lints typed question sets offline, then sends them and prints probabilities, labels and scores as JSON.
CLIbetaRust