User repository on GitHub
Hermes Agent plugin: TypeSafe Jev model routing + trim-then-compress
- Forks
- 0
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Integrations & Adapters
Hermes Jev Router — punkcanyang
A Hermes Agent plugin that adds per-turn model routing — Jev picks cheap, primary or complex before each call — and a trim-then-compress context engine that forbids compressing without trimming first.
- Stars
- 1
- Language
- Python
- License
- MIT
- Reading time
- 2 min
- Published by
- Community
- Author
- punkcanyang
- Last verified
- Oct 2026
Highlights
- Each turn, pre_llm_call asks Jev a Choice — cheap, primary or complex — and rewrites the turn's model; timeouts and low confidence fall back.
- Defaults: min_confidence 0.55, timeout_seconds 8, primary_model deepseek-flash, keep_last_n_turns 6.
- The context engine trim_compress enforces trim then compress; the name must be trim_compress, not the old jev-trim-compress.
- TypeSafe direct only via TYPESAFE_API_KEY; the README forbids Vercel AI Gateway and keeps keys out of the repo.
- Verify with scripts/eval_routing.py --force-live, prove_trim_compress.py, and a $0 fail-closed prove_auto_approve.py.
Quickstart
bash
export TYPESAFE_API_KEY=...
cd /path/to/hermes-jev-router
bash scripts/install_local.sh
hermes plugins doctor ~/.hermes/plugins/jev-routerWatch out
Copy reviewed against the truncated repo README (written in Chinese) on 2026-10-15; the excerpt stops inside the directory listing. The known-limits section states the offline ordering proof uses a deterministic summary for the compress step rather than a full interactive session, and that hooks being visible does not prove enforcement.
Questions about this entry
Short answers drawn from the linked project or page.
A Hermes Agent plugin that adds two things: per-turn model routing where Jev picks a model before each call (cheap for simple turns, stronger for complex, tool-heavy or long-context ones), and a trim-then-compress context engine. The author is @punkcan.
Each turn runs pre_llm_call → a Jev Choice such as cheap / primary / complex → the turn's model is rewritten. Timeouts, exceptions or low confidence fall back to the primary model; min_confidence defaults to 0.55 and timeout_seconds to 8.
The context engine trim_compress forces trim → compress and forbids compressing without trimming first. The engine name must be trim_compress — not the old jev-trim-compress from earlier docs. It can be disabled with JEV_ROUTER_TRIM_ENABLED=false.
With the Hermes CLI installed and a primary model configured, export TYPESAFE_API_KEY, then from the repo run bash scripts/install_local.sh. It symlinks into ~/.hermes/plugins/jev-router, enables the plugin and sets context.engine to trim_compress; verify with hermes plugins list and hermes plugins doctor.
Priority low to high: ~/.hermes/jev-router.yaml, then a jev_router block in config.yaml, then plugins.entries.jev-router.settings, then environment variables. Aliases include enabled = routing_enabled and main_model = primary_model; unrecognized booleans fall back to defaults.
Set JEV_ROUTER_ROUTING_ENABLED=false to always use the primary model, or JEV_ROUTER_TRIM_ENABLED=false to skip trimming while compression can still run. Old aliases HERMES_JEV_ROUTING=0 and HERMES_JEV_TRIM=0 exist; JEV_ROUTER_* wins when both are set.
The README's own list: the offline ordering proof uses a deterministic summary for the compress step; hooks being visible does not prove an enforced full chat; if cheap and primary point at the same model id the comparison still runs but nothing changes; and the router does not hold third-party keys.
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
X: OpenRouter Decision Model Rankings
X: herdr + jev, an agent router
X: A group chat bot with decision room routing
Other listings of this project
Forks and copies of this project in the directory. This entry is the canonical one.
Experimental Hermes plugin that asks Jev to label task complexity, then deterministic policy code picks the cheapest eligible allowed model and returns a routing plan. It does not dispatch the request.
Jev Agents & MCP#community#jevbest#catalog-import
Community
ussyverse/hermes-jev-routerPython
ussyverse/hermes-jev-routerPythonMore Jev resources like this
Effortlane is a local router for Codex on macOS that picks model and reasoning effort with Jev before each user turn, keeps native ChatGPT authentication, and records routing evidence locally. Shadow mode logs recommendations without changing execution.
Jev Integrations & Adapters#community#catalog-import#logicrw
A session-aware, OpenAI-compatible gateway that routes chat completions across provider pools by cost, quality, and capability rules, optionally asks configured decision providers typed choice questions with a deterministic fallback, and records requests, decisions, and outcomes in SQLite.
Jev Integrations & Adapters#community#catalog-import#logicrw
Community
texasoct/jev-gatewayTypeScript
texasoct/jev-gatewayTypeScriptA host-agnostic routing layer that sits between a coding harness and its models, choosing capacity per round from trajectory evidence like tool calls, failures and context pressure, running as a user-scoped daemon and local OpenAI-compatible proxy.
Jev Integrations & Adapters#community#catalog-import#logicrw
Community
vizuh/sabiTypeScript
vizuh/sabiTypeScriptA hook plus /jev skill for Claude Code: Jev picks the cheapest Claude tier (haiku, sonnet, opus, fable) for each prompt and estimates whether the task depends on the conversation, delegating self-contained work to a cheaper subagent.
Jev Integrations & Adapters#community#catalog-import#logicrw
Local tool-routing classifier for coding agents: a gateway and MCP tool that ask Jev which tool to call next, keep your agent's own model and login, and log every decision so you can start by observing.
Jev Integrations & Adapters#community#catalog-import#logicrw
Community
felpsdev/jev-classifierTypeScript
felpsdev/jev-classifierTypeScriptTwo opt-in Hermes Agent plugins that add TypeSafe Jev decision layers: a pre-tool gate for terminal, write_file, and patch calls, and a skill router that names the one skill worth loading before the model call. Stdlib only and fail-open.
Jev Integrations & Adapters#community#catalog-import#logicrw
Community
dogmatiic/hermes-jevPython
dogmatiic/hermes-jevPythonFrom the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.