Why add a decision adapter if I already run a strong LLM?

The LLM keeps writing plans and explanations while Jevbridge returns noul, choice, and score answers for branching. Native Jev is used when a key is set, otherwise your LLM or a heuristic speaks the same protocol.

How do the backend modes differ in practice?

Backends are jev, llm, heuristic, and auto, with auto preferring Jev, then the LLM adapter, then heuristic. Question shapes stay the same across backends.

What do the MCP tools actually expose to my agent host?

Tools are jev_decide for mixed questions, jev_gate for gating computed answers, jev_computer_use for the next GUI action, and jev_recipe for built-ins, plus recipe resources and prompts over stdio.

How do confidence gates handle risky tool calls?

Gate outcomes are execute, confirm, escalate, and abort, so peaked distributions run, middling ones confirm, and destructive clicks do not go unsupervised. ACP proxy mode can intercept upstream tool calls.

Do I need a TypeSafe key to try Jevbridge?

No. It ships a zero-dependency Node 22 library plus MCP and ACP stdio binaries, and the heuristic backend and any OpenAI-compatible LLM speak the same protocol. backend auto uses Jev when a TypeSafe key is present, otherwise the LLM adapter, otherwise the heuristic scorer.

What recipes does Jevbridge ship?

Four: support-route, computer-use, destructive-gate and compaction. Run one from the CLI with node bin/jevbridge.mjs eval destructive-gate, and the MCP server exposes jev_recipe so an agent host can run them too.

How does the ACP proxy mode work?

Set JEVBRIDGE_ACP_UPSTREAM to claude or codex and Jevbridge proxies that upstream ACP agent while intercepting tool calls. Use JEVBRIDGE_ACP_COMMAND for a custom upstream, JEVBRIDGE_INTERCEPT=0 to disable interception, and JEVBRIDGE_SESSION_DIR to move where sessions are stored.

CommunityJev Demos & Experiments

Jevbridge — gamesonrblx

Zero-dependency Node 22 adapter that puts typed Jev decisions beside any LLM, with swappable backends, confidence gates, MCP tools, an ACP sidecar, and offline-ready recipes.

See highlights

Open github.com

Stars: 47
Language: TypeScript
License: MIT
Published by: Community
Author: gamesonrblx
Last verified: Oct 2026

gamesonrblx/Jevbridge

Organization repository on GitHub

View on GitHub

ACP and MCP adapter that bridges TypeSafe Jev with any LLM — computer use and typed decisions alongside Codex, Claude, Grok, and OpenCode.

Forks: 4
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

This is one of several Jev listings for the same project. The canonical entry is Jevbridge — tacticocc.

About this project

[![Contributors][contributors-shield]][contributors-url] [![Forks][forks-shield]][forks-url] [![Stargazers][stars-shield]][stars-url] [![Issues][issues-shield]][issues-url] [![MIT License][license-shield]][license-url] ACP and MCP adapter that bridges TypeSafe Jev with any LLM. Computer use and typed decisions alongside Codex, Claude, Grok, and OpenCode. View Demo · Report Bug · Request Feature. Usage MCP Server ACP Adapter Computer Use Roadmap Contributing License Contact Acknowledgments.

Quoted from the project's README. Read the full README.

Tagscommunity jevbest catalog-import

Works forJev for Classification Jev for Tool calling

Used inJev for Software & DevTools

Languages: TypeScript

Highlights

Evaluates one state against mixed choice, score, and noul questions with swappable backends and unchanged shapes.
Confidence gates map answers to execute, confirm, escalate, or abort for tool calls and computer-use clicks.
Speaks MCP over stdio and ACP so Zed, JetBrains, Claude Desktop, Cursor, Codex, and OpenCode can attach it.
Ships recipes for support routing, destructive gates, context keep and drop, and GUI next-action, runnable offline.

Watch out

Native Jev needs TYPESAFE_API_KEY and LLM fallback needs your own provider key; heuristic numbers are a smoke test because question text leaks into its evidence. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

jev-browser-skill

Standalone CLI skill that drives isolated Playwright Chromium with Jev choosing scoped in-page actions, defaulting to Vercel AI Gateway with TypeSafe as an option.