jev-gate
Pre-flight hook for WorkBuddy and Hermes Agent where Jev classifies each prompt on nine dimensions and injects routing cues.
Overview
Jev Pre-flight Gate runs before the main model reads a prompt in WorkBuddy or Hermes Agent. In about a second, Jev makes a structured judgement of the request and injects the result as a routing hint, for example that a task needs no file reads, can be answered in a sentence or two, or should be acted on without asking for confirmation. The aim is to cut costs from unnecessary tool calls, extra round trips and uncontrolled output length.
The judgement covers task type, whether history or files are needed, whether answers are on disk, whether the prompt is a correction, ambiguity, risk, scope, depth and skill route. The author reports injection costs about 300 to 400 tokens per turn, a net loss of about 1 percent on simple requests that would not read files anyway, but positive on average. It is one hook plus two standard-library Python scripts, with on and off commands, and agent-readable install instructions that require the agent to stop and ask for a key rather than invent one. Documentation is in Chinese.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | What type of task is this prompt? | — |
| Yes / no | Does this request need files or conversation history? | — |
| Choice | What are the ambiguity, risk, scope and depth levels? | — |
How it works
- Hook fires on UserPromptSubmit (WorkBuddy) or pre_llm_call (Hermes)
- Jev judges nine dimensions of the request
- Inject a compact pre-flight block with prohibitive cues
- Main model proceeds with the hints
Strengths
- Avoids unnecessary greps and round trips
- Zero-dependency Python
- Easy on and off switches
Caveats
- Small net cost on trivial requests
- Chinese documentation
- Agents must be restarted after install
Getting started
`git clone https://github.com/adamwdff/jev-gate.git && cd jev-gate && python3 install.py` (or `python3 hermes/install_hermes.py` for Hermes).At a glance
- Type
- Project
- Status
- beta
- Language
- Python
- Runtime
- WorkBuddy and Hermes Agent hooks
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
This project integrates Jev to provide structured decisions for its workflow. See the repository for implementation details.
logicrw/awesome-jev-projects
Frequently asked questions
What does Jev decide?
Nine properties of each prompt, such as task type, whether files are needed, risk and scope.
Does it always save tokens?
No; it costs about 300 to 400 tokens per turn and is net positive on average, per the author.
Which agents are supported?
WorkBuddy and Hermes Agent.
Related projects
Bash-native CLI that filters any stream by meaning with Jev, like grep for semantics, so an agent wakes only for lines that matter.
ProjectbetaTypeScriptTiny JavaScript semantic matcher that classifies text locally in microseconds and escalates ambiguous cases to Jev.
ProjectbetaJavaScriptTransparent proxy that asks Jev to classify each coding-agent prompt and injects the decision into the agent's context before it generates.
ProxybetaPythonOpenClaw plugin where Jev decides whether a group-chat bot should reply and, optionally, which model tier should answer.
PluginbetaTypeScriptOpenCode orchestrator that keeps a cheap sticky parent model and escalates to strong child subagents only when Jev flags a hard turn.
ProjectbetaTypeScriptAdaptive thinking-level router for Pi: Jev or local rules pick low/medium/high/xhigh at task start, on reasoning failures and when stable.
ProjectbetaTypeScript