Agent Router
Pre-release CLI: policy and quota filters, then TypeSafe ranks agent, model, and effort; launches Cursor, Claude Code, Codex, or OpenCode via Herdr.
Overview
Agent Router (github.com/nidhi-singh02/agent-router) is Nidhi Singh's pre-release CLI that reads a task string, applies deterministic rules on enabled models, quotas, and a forty percent reserve on shared accounts, then calls TypeSafe to rank what remains and choose reasoning effort. router run launches Cursor, Claude Code, Codex, or OpenCode inside a Herdr pane via logged-in CLIs rather than raw provider API keys. README marks the project under active development and requires a TypeSafe key with no fallback on every run. Local config and decision history stay under .model-router unless you opt into integrations.
Problem: Developers juggle Cursor, Claude Code, Codex, and OpenCode subscriptions without a quota-aware pick of agent, model, and effort per task.
Built for: Herdr users with multiple agent CLIs logged in who want policy-first eligibility filters then TypeSafe ranking before launch.
First indexed on Jev Directory: 2026-09-24
Creator and team
- Name
- Nidhi Singh
- Handle
- @nidhi-singh02
How Jev is used
- Role in the product flow
- Semantic ranking and effort selection after deterministic subscription eligibility filters
- Primitives
- ChoiceScore
- State in
- Task text plus filtered agent and model candidates after quota and policy rules per README.
- Decision out
- Chosen agent, model, and effort level used to spawn the selected CLI in Herdr.
- Load MODEL_ROUTER_HOME config with enabled models and quotas
- Apply fixed eligibility rules including shared-account reserve
- Send remaining candidates and task text to TypeSafe for ranking
- Pick agent CLI and effort, then launch via Herdr when HERDR_ENV is set
- Record decisions locally under .model-router
Agent Router is the multi-agent launch desk: deterministic code enforces what you are allowed to spend, then Jev picks among legal options. That differs from gargpratyush-jev-router, which focuses on cheapest model routing inside Claude Code only. Pair with dicklesworthstone-skillranker when tasks need skill selection first, or dbreunig-building-with-jev-skill when you are teaching teammates how to shape routing questions. Treat pre-release warnings seriously before pointing shared credentials at router run.
Sourced performance claims
- README states routing always calls TypeSafe with no fallback and rejects recognizable credentials in task text locally.Source: github.com/nidhi-singh02/agent-router README
- README labels the project pre-release and documents Node 20 plus, Herdr pane requirement, and forty percent shared quota reserve.Source: github.com/nidhi-singh02/agent-router README
- Public GitHub repo nidhi-singh02/agent-router had about seventy stars when this listing was drafted.Source: GitHub star count September 2026
Features and stack
Features
- router run task launcher for four agent CLIs
- Quota-aware filters before TypeSafe ranking
- Herdr integration for pane launches
- Local .model-router config and history
- YouTube demo linked from README
Stack
- Node.js 20
- TypeScript monorepo
- Herdr
- TypeSafe Jev
Pricing: Open source pre-release; TypeSafe bills per routing call; agent access uses your existing CLI logins.
Links
FAQ
- Is Agent Router production ready?
- README explicitly marks pre-release active development and asks you to review security notes before real credentials.
- How is this different from gargpratyush-jev-router?
- Agent Router picks among Cursor, Claude Code, Codex, and OpenCode with quota rules and Herdr launch. gargpratyush-jev-router wraps Claude Code or Codex only and routes the cheapest sufficient model per fresh user turn; see its product profile for npm install and status line behavior.
- Can router run without Herdr?
- README states non dry-run launches require running inside a Herdr pane with HERDR_ENV=1.
Related learn guides
Original Jev guidance that pairs with this product pattern.
- Jev use cases
The patterns builders actually search for: moderation, routing, triage, RAG verify, and agent gates.
Related products
Hand-picked neighbors with rich profiles or overlapping tags.