I’ve been using Claude Code for a while and noticed something odd: even simple tasks burned way more tokens than expected. The model is great, but the token usage isn’t optimized — and since they sell API calls, that’s probably not their top priority.
So I built Nexus, an open-source agentic code assistant you can run right in your terminal. It’s lean, local-first, and 3× cheaper to run than Claude Code for the same workflows.
I’ve been using Claude Code for a while and noticed something odd: even simple tasks burned way more tokens than expected. The model is great, but the token usage isn’t optimized — and since they sell API calls, that’s probably not their top priority.
So I built Nexus, an open-source agentic code assistant you can run right in your terminal. It’s lean, local-first, and 3× cheaper to run than Claude Code for the same workflows.