Learn Jev end to end
Free 12-notebook course building 13 agent tools with Jev as the fast decider and an LLM as the slow writer, for under $0.20.
Overview
Learn Jev end to end is a free, hands-on course in 12 short Python notebooks that goes from 'what is Jev?' to building 13 AI tools, including an email triage job, a scam-text detector, a code vulnerability hunter and an agent safety guard. It needs one OpenRouter key, and the README says running the whole course costs less than $0.20.
Its central idea is a fast brain and a slow brain: Jev makes the many small decisions (yes/no, pick one, score), an LLM does the few things that need language, and plain code enforces rules. The README's own runs put Jev at about 0.4 s and $0.02 per thousand decisions against 2.5 to 3 s and $0.08 to $1.67 per thousand for LLMs. Per the listing, a hand-rolled agent loop asks Jev to allow, ask or block before every tool call, with irreversibility and exfiltration checks, sends 'ask' verdicts to a human and fails closed on errors, plus a model router with confidence fallback and an 'am I done?' gate, each measured against labelled fixtures. A playground app and a docs site accompany it.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Should this tool call be allowed, sent to a human, or blocked? | allow · ask · block |
| Yes / no | Is this action irreversible? / Does it exfiltrate data? | — |
| Choice | Which model should handle this request? | — |
| Yes / no | Am I done? | — |
How it works
- Work through the notebooks with one OpenRouter key
- Build tools where Jev makes typed decisions
- Use an LLM only where text is needed
- Gate agent tool calls and route models with confidence fallbacks
- Measure each decision against labelled fixtures
Strengths
- Very low cost to complete
- Covers safety, routing and stopping patterns
- Measured against fixtures, not just demos
Caveats
- Educational code
- Performance numbers are from the author's runs
- Requires an OpenRouter key
Getting started
Follow the 'Start in 3 steps' section of the README or the docs at harshithsunku.github.io/learn-jev-end-to-end.At a glance
- Type
- Project
- Status
- production
- Language
- Python
- Runtime
- Jupyter notebooks and a playground app
- Has code
- yes
- Has tests
- yes
Use cases and integrations
How curated lists describe it
Developer education: a 12-notebook Python course whose hand-rolled agent loop asks Jev a `Choice` (allow / ask / block) with `Noul` irreversibility and exfiltration checks before every tool call, sends ask verdicts to a human and fails closed on errors, and adds a `Choice` model router with a confidence fallback and a `Noul` "am I done?" gate, each measured against labeled fixtures.
yibie/awesome-jevLearn Jev end to end: a free hands-on course. Build 13 AI agent use cases with a fast brain (Jev) and a slow brain (LLM). One OpenRouter key.
hellogumbo/awesome-jev
Frequently asked questions
What will I build?
Thirteen tools, including email triage, a scam-text detector, a vulnerability hunter and an agent guard.
What does it cost?
Under $0.20 for the whole course, per the README.
What key do I need?
One OpenRouter key.
Related projects
Guide and small prototype for using Jev with Herdr to route tasks to coding agents like Claude, Codex and Hermes under code-owned policy.
ProjectprototypeKorean course code for attaching Jev to Claude Code, Codex and OpenClaw: skills, hooks, routers and open-model comparisons.
ProjectproductionJavaScriptCompanion code for the ZazenCodes course on decision models: primitives notebook, use-case examples and a Pi model-router build.
ProjectproductionPythonBeginner manual, video script and three small examples for building an agent harness with Jev via langchain-typesafe.
ProjectprototypePythonChinese translation of 'Jev Engineering for coding agents', design notes on a harness where Jev makes every per-turn decision.
ProjectdiscussionOpenRouter cookbook recipe gating agent tool calls with deterministic checks plus Jev Noul probabilities and fixed thresholds.
Articleproduction