jev-decision-layer
MIT plugin and local MCP server that hands an agent's closed questions to Jev or local Laya, with policy gates and receipts.
Overview
Qualixar Jev Decision Layer lets AI assistants such as Claude, Codex and Copilot hand bounded questions to a decision model: which of several formats fits, whether a draft meets the brief, which team a request belongs to, whether a search result answers the question. It returns one of the given options, a scale position or a yes/no likelihood, each with confidence and a local receipt ID. Every answer is advice; the assistant still decides and asks before acting.
The decision model can be hosted Jev (directly or via OpenRouter) or Laya running locally on a Mac. Nothing is sent until a folder is approved in a private local setup page. Release 1.0.15 includes 20 MCP tools, 55 recipes, 165 synthetic offline fixtures and five host adapters, supported on macOS with Linux experimental. A self-test replays fixtures through the local checking logic without a key, verifying gate contracts but not accuracy.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which of the given task, tool, skill or format options fits? | — |
| Yes / no | Does this draft or result meet the brief? | — |
How it works
- Approve a workspace folder in the local setup page
- Assistant sends a closed question through an MCP tool
- Jev or Laya returns a typed answer with confidence
- Local policy gate checks it and records a receipt; the host keeps execution authority
Strengths
- Nothing leaves the machine until a folder is approved
- Local Laya option on Mac
- Auditable receipts per answer
Caveats
- Verified on macOS only; Linux experimental
- Self-test does not measure provider accuracy or savings
Getting started
Clone the repo and run `plugins/qualixar-jev-decision-layer/scripts/jev selftest` (macOS, Python 3.11+, no key needed).At a glance
- Type
- Agent skill
- Status
- production
- Language
- Python
- Runtime
- Agent plugin and local MCP server
- Has code
- yes
- Has tests
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Open-source decision layer for AI agents. Typed task, tool, skill and review routing through TypeSafe Jev, with local policy gates and auditable receipts.
hellogumbo/awesome-jev
Frequently asked questions
What does it decide?
Closed questions an assistant asks: options, scale positions or yes/no likelihoods about tasks, tools, skills and reviews.
Can it run locally?
Yes, with Laya on a Mac instead of hosted Jev.
Does it act on its own?
No. Answers are advice; the host assistant keeps execution authority.
Related projects
Orchestrator that probes which coding agents are installed and logged in, then has Jev pick the best one for a task.
Agent skillbetaPythonOpen-source Jev skills for Claude Code, Codex, Cursor and Grok: one client to gate, route, guard, verify, triage, review and act.
ProjectbetaTypeScriptLocal gateway, MCP server and Claude Code hooks that let coding agents delegate explicit choices, classification and scoring to Jev.
MCP serverbetaTypeScriptClaude Code and Codex skill that routes a task to one of 27 specialist roles, with an optional Jev engine for catalog selection.
ProjectbetaPythonHooks for Claude Code, Codex and pi that use Jev to suggest a skill each turn, plus a rule gate for subagent model routing.
HookbetaJavaScriptPer-prompt capability router for coding agents that picks at most one installed skill, MCP server, subagent or command via Jev, or none.
MCP serverbetaTypeScript