Clicky's Pro tier caps you at 150 agent messages a month.
That's not a product decision. Agentic desktop tasks cost real money, so the cap is a margin decision.
The obvious fix is routing cheap tasks to cheap models. Here's why nobody does that confidently:
The one to stare at is scope_creep: asked to draft, it sent.
Every other failure a retry fixes. That one's on the user's calendar, in the user's name.
MIT, 47 tests: https://t.co/hTm1wE2axs
Built it for Clicky. Wrote this using @heyclicky@FarzaTV
Clicky's Pro tier caps you at 150 agent messages a month.
That's not a product decision. Agentic desktop tasks cost real money, so the cap is a margin decision.
The obvious fix is routing cheap tasks to cheap models. Here's why nobody does that confidently:
So I built costfloor: per task family, the cheapest model that shows no silent regression against the expensive baseline.
Six structural detectors, no LLM judge. Runs offline, no API key, under a second.
pip install -e . && costfloor demo