How does the plugin pick a model for each message?

Jev reads your typed text and, by default, the last two things you wrote capped at 2,000 characters, then returns a level and an effort from 0 to 3. The level maps to gpt-6-luna, gpt-6-sol or gpt-6-astra on a ChatGPT session, or Haiku, Sonnet, Opus or Fable on Claude.

Why does a short "ok" not move to the small model?

The router treats it as a continuation: the model that made the plan continues the conversation, shown as segue on the status line, because switching models cools the cache and pays for the plan a second time.

What leaves my machine when a message is routed?

Only what Jev needs: your message and, by default, the last two things you wrote capped at 2,000 characters. Agent responses, tool results, file contents and the system prompt never leave, and /jev contexto off sends only the current message.

What happens when Jev is slow or the key is dead?

The router fails open: your message runs on the model your session normally uses and the status line says why. The log stores only numbers — model, tokens, decision and probabilities — never text from you or the agent.

Does it work with an API key or across providers?

No. It is built for subscription sessions and detects API-key or OpenRouter LLM setups and stays disabled, and it also never switches provider, so a ChatGPT session stays on ChatGPT and a Claude session stays on Claude.

What does the savings report actually measure?

report.py reads the local usage log with no network and no LLM and estimates quota: a sample 24-hour run shows 17 messages, 9 model switches, 19 downgraded calls costing $0.09 instead of $0.42, with an on-screen warning not to over-read a sample that small.

How do I install and check it later?

Run the install script, open a new Hermes session and run /jev setup, which asks three questions and stores the key in ~/.hermes/.env; later /jev doctor tests key, provider and version, and /jev stats shows totals since install.

CommunityJev Agents & MCP

Jev Hermes Router

A Hermes plugin that routes every message you type to one of four model levels with Jev deciding in about 300 ms: lightweight for the trivial, maximum for what matters, and continuations stay on the model with the warm cache.

See highlights

Open github.com

Stars: 8
Language: Python
License: MIT
Published by: Community
Author: okjpg
Last verified: Oct 2026

okjpg/jev-hermes-router

User repository on GitHub

View on GitHub

Escolhe o modelo certo pra cada mensagem no Hermes, usando o Jev da TypeSafe. Codex e Claude via assinatura.

Forks: 4
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

Tagscommunity python hermes routing plugin

Works forJev for Routing

Used inJev for Software & DevTools

Languages: Python
Platforms: CLI
Capabilities: Open sourceRequires API key

Highlights

Each message lands on one of four levels — leve, padrão, pesado, máximo — mapping to luna, sol and astra or Haiku, Sonnet, Opus and Fable.
Jev also picks an effort from 0 to 3, applied as reasoning.effort in Codex or a thinking budget in Anthropic.
Continuations like "ok" hold on the model that made the plan so the prompt cache stays warm instead of flipping on every yes.
Risk verbs such as send, delete, publish and deploy never fall to the weakest model, and the minimum effort there is 1.
Jev costs about US$0.00003 per message; the documentation pitches quota saved on subscriptions, not dollars.

Watch out

MIT, but only for Hermes 0.21.4 or newer signed in with a ChatGPT or Claude subscription: it detects API-key and OpenRouter setups and stays off, and it never switches provider. The install script exists to work around a plugin-manager bug in Hermes 0.21.x.

Questions about this entry

Short answers drawn from the linked project or page.

Reactions & coverage

Posts, threads, and videos about this entry from around the web.

Facebook: Intake Router keeps diagnosis out of the answer space at the front desk.

FacebookClinicnotes

YouTube: Jev explained for routing and judging, with a comparison against LLMs.

Hacker News: Vercel on using Jev for defined decisions and GPT-6 Astra when the job also has to write or use tools.

flashbrew4 pointsTypeSafe AI Jev vs. GPT-6 AstraRead the thread on Hacker News

Hacker News: ASTGL replay chose local Laya routing over hosted Jev on one Mac Studio setup.

Jmeg8r15 pointsDoes an open-weight decision model beat a hosted one? Jev vs. LayaRead the thread on Hacker News

Related terms

Glossary definitions related to this entry.

Intent routing

More Jev resources like this

Claude x Jev

A free Claude Code skill that puts Jev in front of the expensive model: thirteen commands sort, check, score, gate and verify with Jev, so Claude reads only the unsure slice and writes the output. Needs python3 and an OpenRouter key.

Jev Agents & MCP#community#python#claude-code

Community

charlesdove977/claude-x-jevPython

Gatekeeper

Claude Code hooks that narrow a roster in YAML, ask Jev which handler owns a prompt, and log the band before anything is blocked.

Jev Agents & MCP#community#python#coding-agent

Community

AgriciDaniel/gatekeeperPython

Jev Harness

An agent harness where Pi runs the loop, Gemini does the work, and Jev makes the small decisions: which model tier to use, which runbook sections to load, whether a tool call may run on a real Postgres database, and whether the report holds up.

Jev Agents & MCP#community#python#agents

Community

PromtEngineer/jev-harnessPython

Codex Sift

A model router for the Codex CLI: on every user turn Jev judges how hard the task is and a YAML policy picks one of three lanes — flash, craft or forge — so premium quota goes to the prompts that need it.

Jev Agents & MCP#community#typescript#codex

Community

romanmeclazcke/codex-siftTypeScript

Jev Router

Launchers for Claude Code, Codex, Grok CLI, and opencode that ask Jev which model-and-effort tier each human turn needs, rewriting only the model field while the real CLI runs unchanged.

Jev Agents & MCP#community#javascript#routing

Community