jev-n8n-patterns
Scaffold of n8n workflow patterns using Jev for PII screening, agent guardrails as hooks, and workflow evals with Jev as judge.
This repository is mostly documentation or write-up rather than runnable code.
Overview
jev-n8n-patterns collects patterns for using Jev inside n8n workflows, in three areas: PII screening on inbound text, guardrails for agents and coding agents implemented as Claude Code and git hooks, and workflow evaluations with Jev as the judge. Everything is designed for stock n8n with one OpenRouter key and synthetic data, under MIT.
The README explains the design principle that code owns the workflow (regex, lookups, counting, date math, redaction and decision tables), while Jev answers only what code cannot, with a threshold tuned on labelled data and a middle band for confirmation or escalation. Planned workflows include a reusable Jev call sub-workflow, a PII screen where code extracts candidates and Jev classifies them, a webhook gated by that screen returning 200, 202 or 422, a log redactor, and an evaluation where Jev scores LLM-drafted support replies and picks send, edit, rewrite or discard. The repository is marked as a scaffold, with most items still planned.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | PII screen policy: allow, redact, quarantine or review? | allow · redact · quarantine · review |
| Yes / no | Does this reply invent facts or leak PII? | yes · no |
| Choice | verdict: what should happen to this drafted reply? | send · edit · rewrite · discard |
How it works
- Code extracts PII-shaped candidates
- Jev classifies candidates and screens text for other categories
- A policy table decides allow, redact, quarantine or review
- Code redacts deterministically and logs an audit row
Strengths
- Clear separation of code logic and model judgment
- Runs on stock n8n with one key
- Cost measured from OpenRouter usage fields
Caveats
- Scaffold: most workflows are planned, not built
- Synthetic data only
At a glance
- Type
- Project
- Status
- prototype
- Runtime
- n8n workflows and hooks
- License
- MIT
Use cases and integrations
How curated lists describe it
Jev (TypeSafe System One) patterns in n8n: PII screening, agent and…
wh000wh000/awesome-jev-live
Frequently asked questions
What will Jev decide in these workflows?
PII classification and policy, guardrail checks for agents, and verdicts on LLM-drafted replies.
Are the workflows ready?
Not yet; the README marks the repository as a scaffold with items planned.
What do I need?
A stock n8n instance and one OpenRouter API key.
Related projects
Helper n8n node for Jev by Brains of Bots: classify, route and score with a probability per answer to send unsure items to review.
ProjectbetaTypeScriptExperiment using Jev through langchain-typesafe as an agent-eval judge, compared with GPT-5.6 and Claude Sonnet 4.6 against human labels.
BenchmarkresearchPythonLangChain post comparing Jev with LLM judges for agent evaluation on accuracy, repeatability, latency and cost.
ArticleresearchPython/TypeScript/Go chatbot safety library checking user input and bot replies with Jev and returning one actionable verdict per message.
LibraryproductionPythonOpen-source agent guardrails that use Jev or LAYA to check prompts, retrieved content, tool calls and outputs for injection and risk.
ProjectbetaPythonUnix-style tools where Jev learns a repo's decision norms and judges each node of a code lead as a bug, lead or clean.
ProjectbetaJavaScript