JevRouter CLI
A local router merges models, subagents, skills and MCP tools into one candidate set and picks who handles the next step.
- 469 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Candidate descriptions and the request go to the external Jev API or OpenRouter.
- Setup changes project instructions and `.gitignore`.
Install
- Works in
Manual install
npx --yes github:BillionsBobby/JevRouter agent start --agent codexThe first Quickstart command in the README installs the skill and project instructions, checks Jev and launches Codex; use `--agent claude` for Claude Code and `--agent cursor` for Cursor.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Candidate descriptions and the request go to the external Jev API or OpenRouter.
- Setup changes project instructions and `.gitignore`.
This is third-party code. Review the repository files before installing.
What it does
JevRouter asks an agent one question: which capability should handle the next step. Models, subagents, skills, MCP tools, CLIs and plugins merge into one candidate set, and the Jev model decides through a typed choice question. Around the answer the router checks availability, permissions, risk level and the need for confirmation. The `agent start` command installs the skill and project instructions, checks the Jev connection and launches Codex, Claude Code or Cursor. Decision receipts are stored locally in `.jevrouter/`.
Who it is for. Developers of agent systems with many tools and subagents who need an auditable choice of executor.
Good fit when
- An agent burns tokens choosing between a dozen tools, and that choice should go to a fast model.
- Before a risky tool runs, the router must check permissions and ask for confirmation.
- Local decision receipts are needed to show why an executor was chosen.
- Candidates can be passed as a list straight to the `route` command, with no registry.
Not a fit when
- Requests cannot be sent to the external Jev API or OpenRouter.
- The agent has two or three tools and the choice is obvious.
- A mature release is required: the project is at 0.1.0.
Example request
Choose which tool should find original sources before summarising from the candidate list, and show the decision receipt.Limitations
It needs Node.js 20 or newer and a key for the official Jev API or OpenRouter (`TYPESAFE_API_KEY`, `JEV_API_KEY` or `OPENROUTER_API_KEY`). Candidate descriptions go to the chosen provider, so keep secrets out of them. The `agent doctor --live` check makes a small paid request. The package is at version 0.1.0, so commands and formats may change. The key stays in the process environment and is not written to project files, while `.jevrouter/` is added to `.gitignore`.
How to disable. Remove the installed skill and the JevRouter instruction block from the project, then the `.jevrouter/` directory. Revoke the Jev and OpenRouter keys at the provider.
FAQ
Which keys work?
A key for the official Jev API (`TYPESAFE_API_KEY` or `JEV_API_KEY`) or an OpenRouter key (`OPENROUTER_API_KEY`). If the variable is not set, the command asks for the key interactively.
How do I verify the setup?
Run `agent doctor`; the `--live` flag adds a connectivity check with a small paid request.
Can I skip a tool registry?
Yes. The `route` command accepts candidates in `--candidates` as a JSON list with a name and description.
Is there an MCP adapter?
Yes, the cookbook has `05-mcp-adapter.md`, next to guides for using it with Codex and Claude Code.
Where are router decisions stored?
Decision receipts live locally in `.jevrouter/`. Setup adds it to `.gitignore` so receipts are not committed by accident.
Discussion
No questions or reviews yet.
Related
CLI toolsA plugin and MCP server that hand no-text decisions to the Jev model and gate shell commands while the LLM writes content
CLI that picks the agent, model and effort for a task by remaining subscription quota and launches it in a Herdr pane.
Jev-Style runs small decision models on your machine: a probability per option, six skills, MCP and a Claude Code guard hook
An MCP server giving an agent three Jev tools: typed yes or no, choice and score questions with probabilities, offline by default.
The Jev classifier as an agent tool: labels items, answers yes or no and ranks candidates with probabilities within milliseconds.
Five skills for the Jev model: sorting records, source checks, rubric scoring and choosing an agent's next action.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| JevRouter (this card) | no | 469 | |
| Agent Router | no | 124 | picks an agent and model by remaining subscription quota; JevRouter picks a tool or subagent through a typed question |
| Intent-Router | no | 95 | turns a vague request into an IntentSpec; JevRouter decides which capability runs the step |
README in short
The Jev model answers one typed choice question, while JevRouter enforces availability, permissions, risk and confirmation around the answer.