Mix any model in one agent team
Use a strong model where it earns its cost and a cheap one everywhere else — in the same team, in one run.
The problem.
One model for the whole pipeline is either overpaying on the easy steps or underpowered on the hard one. Splitting providers usually means standing up separate clients and gluing their outputs together by hand.
How open-multi-agent does it.
Each agent names its own model, and they cooperate inside one team — Anthropic, OpenAI, Gemini, a local Ollama, or any OpenAI-compatible endpoint, mixed freely. One thinking config maps across providers, modelRouting sends work to the right tier, and onTrace spans let you watch per-agent cost and latency instead of guessing.
When this fits.
This fits when different steps have genuinely different difficulty and cost profiles, and you want one team rather than a Python service beside your Node app. If a single model comfortably covers the whole job, you don’t need the mixing.
Taking this to production?
open-multi-agent is MIT-licensed and free to run yourself. When you need it delivered, integrated, or supported on a deadline, 元定义科技 (YuanASI) offers commercial delivery and support.