Does Nerve replace the Hermes task queue and completion state?

It does not. It supervises claims, evidence, budgets, and semantic checks, while cards, dependencies, retries, queues, and done state stay in Hermes Kanban. Completion flows through native kanban_complete after PASS.

What actually happens on a completion request?

Mechanical criteria are proved locally; Jev judges only the remaining semantic items. Non-PASS leaves the card running with the missing criteria listed, and deterministic FAIL cannot be semantically overridden.

How do you keep Jev token overhead down?

Workers start with zero Jev tool schemas, healthy work stays silent, and a failure or checkpoint may trigger one Jev call whose directive appends once to the next tool result. The target is tiny fixed overhead on healthy runs.

Can I run local Laya decisions instead of hosted Jev?

Jev stays authoritative while paired Laya decisions log alongside, or the local Laya sidecar decides with LOCAL_ONLY receipts. Shadow failures never change the authoritative result.

Which release is stable, and what is on main right now?

Current main identifies as 0.3.2.dev0 while the latest stable release is v0.3.1, and Hermes catalog installs follow exact reviewed commit-SHA pins that may lag main.

What triggers a Jev call during a supervised Kanban worker?

The local ROI router stays silent on healthy work; only repeated failure or an unhealthy 40% or 70% checkpoint may trigger one Jev call.

Does the plugin pull in model runtimes like torch?

No. The core plugin is dependency-free, and the optional Laya runtime lives in a separate process started with python -m hermes_nerve.reflex.laya_service.

CommunityJev Agents & MCP

Hermes Jev

Async supervisory layer for Hermes Kanban workers: locked DoD contracts, evidence tracking, budgeted trajectory checks, and Jev only for unresolved semantic criteria.

See highlights

Open github.com

Stars: 38
Language: Python
License: MIT
Published by: Community
Author: keeltrace
Last verified: Oct 2026

keeltrace/hermes-jev

User repository on GitHub

View on GitHub

Nerve is a supervisory nervous system for Hermes agents, adding typed System One decisions, ranking, verification, token-aware oversight, and an opt-in tool gate powered by TypeSafe Jev or Open Source Laya

Forks: 11
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

Nerve is an asynchronous System-1 supervisory layer for Hermes Agent. 0.3.1 is the stable maintenance release for the profile-aware 0.3.0 line: it concentrates bug fixes, hardening, completion/retry integrity, configuration correctness, setup QOL, and the explicitly approved opt-in ask-only Action Gate while preserving v0.2.3-compatible Legacy behavior when no profile is selected. Other 0.3.2 feature work remains on the development line. Nerve/Reflex remains a watchdog and forecaster; the main Hermes…

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

Tagscommunity jevbest catalog-import

Works forJev for Classification Jev for Ranking Jev for Tool calling

Used inJev for Software & DevTools

Languages: Python

Highlights

Locked DoD plus evidence states separate claims, observations, and verified passes across local and SSH workers.
Deterministic verifiers cover mechanical completion; semantic remainder batches into bounded Jev verification.
Trajectory supervision uses CONTINUE, WATCH, REPLAN, and BLOCK with token budgets and failure fingerprints.
Controller-owned probes prove retry and redispatch idempotency without touching the workspace or network.

Watch out

Jev never owns canonical cards, runs, or completion state, and current main is development past the stable release with catalog pins lagging until reviewed. 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-judge-mcp

MCP server over stdio that gives coding agents fourteen Jev judgment tools, turning evidence plus an enumerable question into auto, review, or escalate actions from calibrated probabilities.