jev-subagent-router
Routes each subagent task to the cheapest model smart enough for it, using Jev to rate task difficulty from low to max.
Overview
jev-subagent-router stops agent setups from spending their most expensive model on trivial tasks like summarising logs. Jev (as typesafe/jev-latest via OpenRouter) reads the task and rates its difficulty from low to max with probabilities in about 0.3 seconds. Each level maps to a minimum Artificial Analysis intelligence score, and the router only raises a level when Jev is confident; risky topics such as data loss, security or real money are floored at high.
Routes are tried free first: Antigravity agy, then OpenAI codex subscription quota, then Claude Code subagents. Within a route the cheapest model wins, then the lowest reasoning effort. A weekly job updates the CLIs, re-lists every model and reasoning level and re-scores them, and the Claude Code status line shows 5-hour and weekly quota. It is aimed at developers who juggle several agent subscriptions. Some file names and output are in Turkish.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Score | How difficult is this task? | low · ... · max |
How it works
- Task text sent to Jev for a difficulty rating
- Level maps to a minimum intelligence score; risky topics floored at high
- Pick cheapest qualifying model per route: agy, then codex, then claude
- On exhausted quota, mark the route and fall through; refresh catalog weekly
Strengths
- README reports about 0.3 s per routing decision for a fraction of a cent
- Uses free and subscription quota before paid routes
- Self-updating model catalog
Caveats
- Depends on third-party CLIs and intelligence scores
- Difficulty-to-score mapping is hand-set
- Partly Turkish naming
Getting started
python3 secici.py "<task description>"At a glance
- Type
- Project
- Status
- prototype
- Language
- Python
- Runtime
- CLI / Claude Code hook
- Has code
- yes
Use cases and integrations
How curated lists describe it
Route every AI subagent task to the cheapest model smart enough for it — agy…
wh000wh000/awesome-jev-live
Frequently asked questions
What does Jev decide?
The difficulty level of a task, which sets the minimum model intelligence required.
Which routes does it use?
Antigravity agy, then OpenAI codex, then Claude Code subagents, cheapest qualifying model first.
How does it stay current?
A weekly job re-lists and re-scores every model and reasoning level.
Related projects
- jev-router★ 593
Per-turn model routing for Claude Code and Codex: wraps the real CLI and lets Jev send simple turns to fast models and hard ones to strong.
ProxybetaJavaScript TypeScript/Bun launcher that asks Jev which model and reasoning effort each Codex, Claude Code or Pi turn should use.
ProjectbetaTypeScriptGo CLI model router that uses one Jev analysis to recommend a model, harness and effort level, then selects deterministically.
CLIbetaGoClaude Code and Codex plugin where Jev judges each turn for model routing, delegation, evidence search and compaction.
ProjectbetaTypeScriptClaude Code, Claude Desktop and Codex plugin where Jev reads a task and ranks live models and reasoning effort by quality, speed and cost.
PluginbetaTypeScriptLocal pass-through proxy for Claude Code and Codex where Jev picks a model tier per human turn and only moves up within a session.
ProxybetaTypeScript