June 26, 2026
Route now, pay later?
Show HN: Smart model routing directly in Claude, Codex and Cursor
One app to pick your AI? Fans are curious, critics say it could cost more, not less
TLDR: Weave launched a tool that automatically picks which AI chatbot should answer each coding request, aiming to make apps like Claude, Codex, and Cursor smarter with one setup. Commenters weren’t sold: the big argument was whether all this switching saves cash or secretly makes bills worse.
A new Show HN launch is promising a very seductive idea: point your coding assistant at one local tool, and it will quietly choose the “best” artificial intelligence model for each request. In plain English, it’s like a switchboard for chatbots, connecting Claude, OpenAI, Gemini, and a pile of open-source options while bragging that it’s the top router on a public leaderboard. The pitch is all convenience: one setup command, your own keys stay on your machine, and a dashboard lets you watch where your requests go.
But the real show in the comments was the rising panic over one thing: money. Several readers immediately pounced on the same fear — if the tool keeps bouncing between different AI systems, doesn’t that wreck caching, the money-saving trick where repeated work gets reused? One commenter basically said, “Cool story, but how does this not become a budget bonfire?” Another warned that smaller, cheaper models can glitch out, stop early, or get stuck in weird loops, turning “smart routing” into “smart chaos.”
Then came the comparison drama. One person cut straight to the obvious question: how is this different from Cursor’s ‘auto’ mode? And in the most Hacker News-coded twist possible, another builder casually dropped in: “It is funny. We are building something similar.” Nothing says startup season like announcing your shiny new invention and instantly finding out three other people are already in your lane.
Key Points
- •Weave launched a model-routing proxy that selects a model per request across Anthropic, OpenAI, Gemini, and OpenAI-compatible endpoints.
- •The product is designed to work with developer tools including Claude Code, Codex, Cursor, and opencode through a local or hosted endpoint.
- •The router supports Anthropic Messages, OpenAI Chat Completions, Gemini native APIs, streaming, tools, vision, and additional model families via OpenRouter.
- •A quickstart installer is provided through `npx @workweave/router`, while a self-hosted setup can be run locally with Postgres and a dashboard on port 8080.
- •The article cites a RouterArena Acc-Cost Arena score of 76.09 and describes the routing logic as using an on-box embedder and a scorer derived from Avengers-Pro 2.