Skip to content
Metatext Decide — typed decisions for your agents, with a probability on every answerGet early access
Metatext
Sign Up

jev-fuse

Governance proxy between Jev or local Laya and agents that turns probabilities into allow, ask, deny, keep, truncate or drop actions.

MCP serverbetaApache-2.0
View on GitHub ↗

Overview

JEV Fuse is a governance runtime, safety gate and audit plane that sits as a reverse proxy between callers (coding agents, services, CLIs and SDKs) and decision engines such as TypeSafe Jev or local Laya. It translates raw probabilities into deterministic policy actions: ALLOW, ASK, DENY, KEEP, TRUNCATE or DROP.

The README describes the problems it targets: naive 0.5 cutoffs producing harmful edge-case actions, agents either prompting for every harmless read or failing open on dangerous commands, shell constructs (find -exec, subshells, backticks, redirects) that bypass regex matching, bursty duplicate evaluations exhausting quotas, and no tamper-evident record of why something was allowed or blocked. It integrates with Claude Code, MCP and AI SDKs, per the list summaries.

What the decision model decides

TypeQuestionAnswers
ChoiceShould this action be allowed, asked about or denied?ALLOW · ASK · DENY
ChoiceShould this context be kept, truncated or dropped?KEEP · TRUNCATE · DROP

How it works

  1. Caller sends a decision request through the proxy
  2. Fuse forwards to Jev or Laya and receives probabilities
  3. Policy maps probabilities to a deterministic action with fail-closed bounds
  4. Record the decision in an audit log

Strengths

  • Avoids razor-thin threshold decisions
  • Shell-aware command analysis
  • Deduplicates bursty evaluations and keeps an audit trail

Caveats

  • Additional proxy to operate
  • Benefit depends on policy configuration

At a glance

Type
MCP server
Status
beta
Runtime
Reverse proxy with MCP integration
Has code
yes
License
Apache-2.0

Use cases and integrations

How curated lists describe it

  • A governed execution layer between typed-decision models (TypeSafe Jev) and agent/AI tooling like Claude Code, MCP, and AI SDKs, turning probabilistic decisions into deterministic, policy-controlled, and auditable actions.

    logicrw/awesome-jev-projects

  • A governed execution layer between typed-decision models (TypeSafe Jev) and agent/AI tooling like Claude Code, MCP, and AI SDKs, turning probabilistic decisions into deterministic, policy-controlled, and auditable actions.

    hellogumbo/awesome-jev

Frequently asked questions

What does Fuse decide?

It turns decision-model probabilities into actions such as allow, ask, deny, keep, truncate or drop.

Which engines does it front?

TypeSafe Jev and local Laya.

Why not use a 0.5 threshold?

The README argues edge cases near 0.5 need fail-closed bounds.

Related projects

  • claude-jev (darwintechlab)

    OpenJev for Claude Code: plugin and MCP server giving Claude jev_choice, jev_noul, jev_score and jev_ask tools on the live API.

    MCP serverbetaTypeScript
  • typesafe-mcp (MarkChu-git)

    Bun-based MCP server wrapping Jev with compact answers and a code-computed act, review or abstain verdict.

    MCP serverbetaTypeScript
  • actiongate-jev

    Gateway that checks agent tool calls with policy plus Jev evidence, then binds approval to a single-use permit for that exact action.

    MCP serverbetaTypeScript
  • hermes-jev (DoGMaTiiC)

    Two stdlib-only Hermes Agent plugins: a Jev pre-tool gate with a jev_ask tool, and a skill router that names one skill per turn.

    Agent skillbetaPython
  • jev-agent-authorization

    Starter kit that gates every MCP tool call with Kinde identity checks plus Jev risk signals, then allows, holds for approval or stops it.

    MCP serverbetaTypeScript
  • jevguard-mcp

    Zero-dependency Python MCP server where Jev gates shell commands, audits patches and resolves decisions, with SQLite caching.

    MCP serverbetaPython