Alder Mini
Proposed first route: @cf/openai/gpt-oss-20b through Cloudflare Workers AI. Alder would hold the provider token on its server and check your Alder account allowance. You would not need a Cloudflare account or a separate model app.
Cloudflare publishes 10,000 free Neurons per account per day, shared by every creator using this route. It is a finite pool, and a coding task can make several calls. We will test the full Studio tool loop and confirm provider terms for Alder's shared end-user use before invitations. No payment method or paid fallback is part of this pilot.
Cloudflare says it does not use Workers AI customer content to train models absent explicit consent. Alder would still send the context needed for each model request; privacy details must be published before access opens.
Cloudflare model details ↗ · Published free allowance ↗ · Data use ↗
Groq GPT-OSS 120B
We evaluated a openai/gpt-oss-120b adapter through Groq. Groq's free organization allowance is shared and has an 8,000-token-per-minute cap. Our captured Studio agent request already exceeds that cap, so the current gateway prototype refuses Groq for model listing, usage, and completions. Alder will not silently switch between providers.
Groq free limits ↗ · Model details ↗
Show the route as it really works.
The desktop Models page will display the active upstream reported by Alder's authenticated gateway. We will publish actual account and shared limits after a signed-in plan, approved edit, playtest and restore run. Neither pilot is deployed or available to creators yet.
OpenCode's promotional free models were evaluated, but their published terms and temporary availability do not establish a shared Alder launch route. The development computer's local model is not part of the creator offer. Creator builds do not route through Claude Code or Antigravity CLI.