Open-source auto-router prices cache-aware model routing for Claude Code and Cursor
airesearch12 · x · 2026-09-21
A developer ships an update to the Jev auto-router: local open-source model support and reuse of your own Claude Code / Codex subscription. Each request is classified in 0.5s, every candidate model is priced for that exact request including redo risk, and the cheapest expected total wins; cheap answers get checked before you see them. Key insight: warm input cache bills the next turn at roughly a tenth of the price, so the real question per turn isn't 'which model is best' but whether a better model is worth losing the cache. MIT-licensed, sits in front of Claude Code, Codex, opencode or Cursor, reserving your plan's limits for hard work.
More from coding & agent
- Rebuilding a 2010 round-the-world trip with Claude from 1,300 tweets and 11,000 photos — jozuejosh · 2026-09-21
- Dev adds a perf monitor to jev: agent watches bottlenecks and flips settings every 5 seconds — tomjohndesign · 2026-09-21
- Watching Claude use a Charuco board to estimate product dimensions from a photo and model it in OpenSCAD — _Stocko_ · 2026-09-21
- AI intern cuts job boot time ~60%, saving ~5 CPU-days per day — DanielLockyer · 2026-09-21
- Anthropic engineer demos 22-minute async Claude Code workflow that runs unattended — anthara_ai · 2026-09-21
- Would coding agents do fine with just bash? Benchmarking without read/write/edit tools — lucasmeijer · 2026-09-21