ASI:One
The idea worth taking seriously is routing: rather than one model trying to do everything, ASI:One plans the task and delegates steps to agents published on Agentverse, so capability grows with the marketplace rather than with the next training run. Making it OpenAI-compatible is a smart adoption shortcut — your existing client works after a base-URL change. What is missing is commercial clarity: no price list, no rate card, and a consumer pitch that promises far more than the developer docs demonstrate.
Pros
- OpenAI-compatible API, so existing SDKs, LangChain and MCP-era tooling connect with a base-URL swap
- Three model sizes — asi1-mini, asi1 and asi1-ultra — behind one interface
- Models plan multi-step work and delegate to specialist agents rather than answering alone
- Backed by Fetch.ai and tied into its existing Agentverse agent marketplace
- Documented integration paths for Claude Code, Cursor and LangChain
Cons
- No pricing of any kind is published — not per token, not per plan
- The marketing site renders as an empty JavaScript shell to anything but a browser
- Several documented doc paths return 404, so the reference material is incomplete
- Answer quality depends on third-party agents on Agentverse, whose reliability is not vouched for
- No country or legal entity for Fetch.ai is stated on either site