jev-research
Guide and small prototype for using Jev with Herdr to route tasks to coding agents like Claude, Codex and Hermes under code-owned policy.
This repository is mostly documentation or write-up rather than runnable code.
Overview
Jev + Herdr: Practical AI Agent Orchestration is a public guide to combining TypeSafe's Jev with Herdr, which manages workspaces, panes, processes and coding agents, and agents such as Claude, Codex and Hermes. It explains a reusable architecture, includes a small Jev routing prototype, and reviews a browser worker built with Jev.
The division of labour is: Jev makes small typed judgments (route a task, score risk, select evidence, decide whether more context is needed); Herdr runs and tracks the workers; agents investigate, implement and test bounded tasks; coordinator code owns policy, thresholds, audit records, retries and irreversible actions; and people approve high-impact or ambiguous work. The README stresses constraining Jev's answers to an allowlist so they never become shell text, deployment commands or merge authorisation, and gives example questions such as choosing among claude, codex, hermes, deterministic or human.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which worker should handle this task? | claude · codex · hermes · deterministic · human |
| Score | How risky is the next step? | read-only · reversible · risky · destructive |
| Yes / no | Does this require human approval? | yes · no |
How it works
- Snapshot task, repository, policy and available workers
- Jev judges route, risk, readiness and need for approval
- Coordinator policy checks the result
- Herdr prompts the chosen agent
- Jev reviews output and picks the next route
Strengths
- Clear separation of judgment, policy and execution
- Allowlisted answers prevent unsafe actions
- Applicable beyond the author's setup
Caveats
- Mostly a guide with a small prototype
- Requires Herdr and a TypeSafe key
At a glance
- Type
- Project
- Status
- prototype
- Runtime
- Guide plus prototype
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
A Jev and Herdr integration guide with a prototype for routing tasks to different Agents.
logicrw/awesome-jev-projects
Frequently asked questions
What does Jev decide?
Which worker gets a task, how risky the next step is, and whether human approval is needed.
What is Herdr's role?
It manages agent workspaces and processes and reports their state.
Can Jev run commands?
No. Its answers are limited to an allowlist and never become shell text.
Related projects
Prompt router for Claude Code, Codex and Antigravity that picks model, effort, skill and language per request via Jev.
ProjectbetaPythonFree 12-notebook course building 13 agent tools with Jev as the fast decider and an LLM as the slow writer, for under $0.20.
ProjectproductionPythonKorean course code for attaching Jev to Claude Code, Codex and OpenClaw: skills, hooks, routers and open-model comparisons.
ProjectproductionJavaScriptCursor plugin whose hooks ask Jev whether subagent spawns, file reads, writes and shell commands are worth taking.
PluginbetaJavaScriptLocal Laya decision layer for DeepSeek Harness: routes prompts, gates risky tool calls and swaps models, all offline at no token cost.
ProjectbetaLocal-first macOS coding workspace that runs your agents and lets a local Laya or hosted Jev selector pick a route the host validates.
ServicebetaRust