I read about Fusion by @OpenRouter a few weeks ago and got obsessed since it claimed it was able to achieve a higher level of intelligence than a single frontier model.
Their approach was simple: when you prompt, the same prompt goes to a panel of frontier models and then to a judge which compares the answers, and finally a synthesizer which writes the final response from the consensus, gaps, and contradictions. According to their benchmarks, this approach got really good results.
So I just had to try it and honestly, the outputs were solid, but every prompt runs the whole council on their API credits. It got expensive fast.
Problem was, like most of you I already pay for Claude Code and Codex, and my keys live across @Vercel AI Gateway and OpenRouter anyway. And, I just wanted a UI to configure the council instead of setting it up via code.
So I built OpenFusion.
Your own model council, running on the subscriptions you already pay for, behind one endpoint. It's a local canvas where you build the whole thing: add nodes, pick the source + model, set each one as panel / judge / synthesizer, toggle web, and the graph becomes one OpenAI-compatible endpoint you can use as a single model inside your AI agents.
Here's a video of it working through the council with one prompt and @AnthropicAI's Fable as the judge (repo below ๐)
a clear evolution from of a project from trying to do 0-to-1 when what the world needs more is 1-to-100
study us
learn from us
next chapter now on solana. honestly what really matters is what gets built and shipped using @0xCapx
chain, network, eco, protocol are secondary!
๐จ๐ณ NOW: China's Tencent begins testing an AI assistant called Xiaowei on WeChat, allowing select users to interact via text or voice, per Bloomberg.
Soon we'll have endless agents that can write, code, design, research, and support.
The scarce thing becomes coordination.
Who decides what gets done, in what order, by whom, and against what standard?
That's where the leverage moves.