(1) Cache the non logged-in pages. You can even host static pages in an unmetered way. They maintain their web service software. They should just do that. If an account is knocking over infra by scraping, ban the account. Make it slightly harder to sign up or something, if necessary.

(2) Vibe coding agents like Claude Code, Codex, Copilot and the grok thing all prefer local copies of a project to scraping web services. Why would a developer using LLMs be hammering their own codeberg site?

(3) If they are worried about AI companies shifting costs (What costs though? See point 1), they should block training scrapers, or - even better - offer a service that resells zero-log-retention / enclaved open-weight LLM inference. This cuts off the revenue stream to the AI companies, and probably lets codeberg skim a bit to expand their own services.

We used to use github at work. It was falling over well before cursor was a thing. Blaming AI for their 1-2 nines of availability is... questionable.