Agent Evaluation 最佳实践
综合自: Langfuse, Opik, Bloom, 以及 agent evaluation 相关学习笔记
6 篇文档带有标签「最佳实践」
经过验证的最佳实践
查看所有标签History-Item-as-Checkpointer: Context-Lossless Design
Scope: Designing a robust state management system where the dialogue history itself serves as a fine-grained checkpointing mechanism for error recovery and state restoration.
LLM Error Handling & Retry: Best Practices Design
Scope: Structured error classification, automatic retry mechanisms, fallback strategies, and recovery patterns for LLM client libraries and agent frameworks.
Pull + Debounced Push 混合流式设计
设计来源: Kimi (kosong) + Pydantic-AI 融合
Rust Actor Runtime 作为 Agent Harness 基座：Kameo vs Ractor
综合自: kameo, ractor
机器人情绪与动作系统设计最佳实践
综合自：reachy-mini-conversation-app（Pollen Robotics）