Point tabane at the coding CLIs you already pay for (claude · codex · gemini) and it routes each task to the best-fit agent, then reroutes before any rate-limit window runs out. Plug in any CLI you can wrap, local models too (qwen · deepseek · glm). Bring your own compute. Free & open source.
go install github.com/eklavyagoyal/tabane/cmd/tabane@latest