Unverified50% confidenceFactExact time
MoA 的工作机制是指定几个模型作为顾问（reference）各自独立作答，再由一个更强的模型作为决策者（aggregator）综合判断给出最终结果
1
Sources
50%
Confidence
Long-term
Relevance
7/15/2026
First Seen
Sources
Related Claims
Unverified配置代码专用MoA时选用代码专项模型作为顾问、通用模型作为决策者的组合优于通才顾问加通才决策者的组合77% similarUnverifiedEntering /model in the conversation interface, selecting Mixture of Agents, and choosing the corresponding preset allows the entire conversation to continuously follow the MoA workflow.72% similarUnverifiedThe interactive configuration process for selecting an advisor model and a decision-maker model can be launched via the command hermes moa configure71% similarUnverifiedMoA 相比 AutoGen、LangGraph 等框架更轻量，不追求Agent自主规划与工具调用，专注于多视角答案融合66% similarUnverifiedIn Agent scenarios, evaluation must assess not only the final output but also the reasonableness of intermediate decision paths.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/514748API
curl https://kongchang.com/api/v1/knowledge/claims/514748MCP
get_claim(id=514748)