Hey Product Hunt! I'm Florian, and I made Fewr.

If you use Claude Code or Codex every day, you've probably hit your plan's limits. I measured where that usage goes, and found something counterintuitive: raw tokens aren't usage. Plans weigh tokens by type and model, most of a session is cheap cached context, so trimming it barely helps. What costs is which model writes the code.

So Fewr splits the work. Your model stays the lead and keeps architecture, design and security calls. Lookups, tests and well-specified code go to cheaper models (Haiku or Sonnet, Luna or Sol on Codex) at the right effort.

In benchmarks with the same model and effort on both sides, graded by hidden checks and measured as real cost (not tokens): 

- Up to 53% less on Codex
- Up to 43% on 5 Claude Code tasks

👉 One command: `npx @fewr/cli`
👉 Type `/fewr` in any chat to see what it saved
👉 Runs on your machine, no server, no telemetry, free and open source

I'd love your feedback, especially on tasks where you think delegation got it wrong.

Hey Product Hunt! I'm Florian, and I made Fewr.

If you use Claude Code or Codex every day, you've probably hit your plan's limits. I measured where that usage goes, and found something counterintuitive: raw tokens aren't usage. Plans weigh tokens by type and model, most of a session is cheap cached context, so trimming it barely helps. What costs is which model writes the code.

So Fewr splits the work. Your model stays the lead and keeps architecture, design and security calls. Lookups, tests and well-specified code go to cheaper models (Haiku or Sonnet, Luna or Sol on Codex) at the right effort.

In benchmarks with the same model and effort on both sides, graded by hidden checks and measured as real cost (not tokens): 

- Up to 53% less on Codex
- Up to 43% on 5 Claude Code tasks

👉 One command: `npx @fewr/cli`
👉 Type `/fewr` in any chat to see what it saved
👉 Runs on your machine, no server, no telemetry, free and open source

I'd love your feedback, especially on tasks where you think delegation got it wrong.

Fewr

Save up to 53% less Claude Code and Codex usage

Save up to 53% less Claude Code and Codex usage