Case study · located work
jev-router
An npm-published CLI wrapper that launches Claude Code or OpenAI Codex through a loopback proxy, with a documented test suite covering shared policy, both request formats, and decision display.
Observedgargpratyush
- Problem
- An npm-published CLI wrapper that launches Claude Code or OpenAI Codex through a loopback proxy, with a documented test suite covering shared policy, both request formats, and decision display.
- Jev’s role
- One Jev call per fresh user turn returns task-complexity signals that pick an abstract model tier (fast/balanced/strong/long); an explicit local policy layer (explicit overrides, fail-open on error, no downgrade at low confidence) turns that into the concrete model.
- Architecture
- User turn → loopback proxy → Jev tier recommendation → src/policy.mjs deterministic rules → concrete Claude/Codex model → forwarded request.
- Why it matters
- Routing is explicitly fail-open and confidence-aware in the policy code, not in Jev itself — the README states Jev failure never blocks the CLI and low confidence never triggers a downgrade.
Code boundary
Jev answers the bounded questions described above. Thresholds, retries, side effects, and anything that must be reproducible stay in the surrounding application code. The research record does not capture this project’s exact question set or threshold values; read the linked source before copying the architecture.
Sources
Limitations
- A located repository or demonstration does not establish production reliability or independent validation.
Reusable pattern and related records
- Cascade router PlausibleA cheap decision chooses whether to use rules, a small model, a premium model, a specialist, or a human.
- Agent Model Router Authored HypothesisA drop-in router that selects model, reasoning effort, and fallback policy per step.
- The strongest near-term architecture is Jev as a complement and control layer around generative models. PlausibleJev produces decisions rather than prose, and Vercel exposes it through an evaluation-oriented API.
- Confidence gate PlausibleAutomation proceeds above a threshold; ambiguous cases go to another check or a person.