QarthO

Elysia AI doc search "Arona" cost ~$27/month It's basically a standard RAG $12 - Hetzner CPX32, a bit overkill an could be reduced $15 - Groq GPT OSS 120B with ~700 requests/day The backend is built with Elysia (obviously) Everything is hosted on a single Hetzner VPS with Coolify proxied by Cloudflare Standard Postgres 18 with pg_vector extension is sufficient enough for our documentation site Every 6 hours, it get Elysia documentation markdown and split each pages into a small chunk by title level and indexed in sequential order Each chunk then gets its own embedding that can be retrieved by vector search Each search retrieve its nearby chunks as well for relevant information We then gave AI tools for either search, read the entire pages or get a table of contents, then it ran in loop until satisfied The good thing is VPS is a fixed cost and CPX32 is a bit overkills for our case that we intentionally left because we are planning to use it with other related Elysia documentation projects as well We are using GPT OSS 120B because it's cheap and performant enough As for Groq, the cost could be significantly reduced by tuning our RAG to reduce amount contents or more aggresive content summarization similar to SKILLS Overall, could be cheaper but currently that's how our AI search works

T3 Code is now available for everyone to use. Fully open source. Built on top of the Codex CLI, so you can bring your existing Codex subscription.