Show HN: Smart model routing directly in Claude, Codex and Cursor
Points and comments are a snapshot, not live.
A drop-in proxy routes each AI request to the optimal model in <50ms, cutting costs by 40-70%.
Weave Router is a local proxy for Anthropic, OpenAI, and Gemini that selects the best model per request using an on-box embedder. It supports streaming, tools, vision, and OSS models via OpenRouter. Costs are cut 40-70% by routing simpler tasks to cheaper models. The router is cache-aware to avoid increased costs from context misses. One-line install: `npx @workweave/router`. Self-hosting also available.
What commenters are saying
Several commenters questioned cost savings due to cache misses when switching models. The developer responded that the router is cache-aware, raising the threshold to switch models when cache is warm. Others noted that in practice, routing often stays within 1-3 models per agent loop. A key distinction: the router can use subscription billing (Claude Pro/Codex) when available, falling back to API billing.