- The report describes a production system combining agents, tools and industry context.
- Company-reported routing figures are not independently verified performance measurements.
Beyond model access
A November 3 QbitAI analysis examines how Chinese company 海马云 is building RunningHub around computing, workflow orchestration and production data, rather than access to models alone. Founder and chief executive 党劲峰 argues that vertical production tasks offer a more viable business than generic consumer AI.
The report describes a three-part orchestration system: agents interpret requests, break tasks down, call tools and check results; a toolchain includes ComfyUI workflows, LoRA training, cloud desktops and professional software; industry context supplies specialist production knowledge.
Routing and professional workflows
According to the company, about 60% of RunningHub model calls already use intelligent dynamic routing. Agents distribute simpler tasks to smaller models and more complex ones to larger models, while considering the user's intended output.
The report also describes Blender running on edge-hosted cloud desktops, allowing users to hand an engineering project back to manual control for detailed changes. These are reported platform capabilities and company claims, not independent evidence of cost savings or production quality.
Sources & context
Go to the original material. Company claims remain attributed to their sources.
01
