arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2601.02577cs.AIastro-ph.IMhep-ph
收藏

Orchestral AI: 一个用于智能体协调的框架

Orchestral AI: A Framework for Agent Orchestration

Orchestral AI

Alexander Roman, Jacob Roman

AI总结:

Orchestral AI 提供了一个统一的Python框架,用于跨多个LLM供应商构建智能体,通过类型安全接口和自动工具模式生成,简化了工具调用和跨供应商集成。

摘要:

随着LLM智能体框架的迅速普及,开发者不得不在通过供应商特定的SDK实现供应商锁定和复杂的多包生态系统之间做出选择,后者会掩盖控制流并阻碍可重复性。由于碎片化的API、不兼容的消息格式以及不一致的流和工具调用行为,跨多个LLM供应商整合工具调用仍然是核心的工程挑战,这使得构建便携、可靠的智能体系统变得困难。我们引入了Orchestral,一个轻量级的Python框架,它提供了一个统一的、类型安全的接口,用于在主要供应商上构建LLM智能体,同时保持科学计算和生产部署所需的简洁性。Orchestral定义了一种单一的通用表示法,用于消息、工具和LLM使用,能够在不同供应商之间无缝运行,消除手动格式转换并减少框架带来的复杂性。自动从Python类型提示生成工具模式规范消除了手动编写描述符的需要,同时在供应商之间保持类型安全性。同步执行模型支持流式传输,使行为确定性、调试简便且实时交互成为可能,而无需引入服务器依赖。该框架的模块化架构清晰地将供应商集成、工具执行、对话协调和用户界面分离,使扩展无需架构纠缠。Orchestral支持在较大框架中发现的高级智能体功能,包括丰富的工具调用、上下文压缩、工作区沙箱、用户批准工作流、子智能体、内存管理以及MCP集成。

英文摘要:

The rapid proliferation of LLM agent frameworks has forced developers to choose between vendor lock-in through provider-specific SDKs and complex multi-package ecosystems that obscure control flow and hinder reproducibility. Integrating tool calling across multiple LLM providers remains a core engineering challenge due to fragmented APIs, incompatible message formats, and inconsistent streaming and tool-calling behavior, making it difficult to build portable, reliable agent systems. We introduce Orchestral, a lightweight Python framework that provides a unified, type-safe interface for building LLM agents across major providers while preserving the simplicity required for scientific computing and production deployment. Orchestral defines a single universal representation for messages, tools, and LLM usage that operates seamlessly across providers, eliminating manual format translation and reducing framework-induced complexity. Automatic tool schema generation from Python type hints removes the need for handwritten descriptors while maintaining type safety across provider boundaries. A synchronous execution model with streaming support enables deterministic behavior, straightforward debugging, and real-time interaction without introducing server dependencies. The framework's modular architecture cleanly separates provider integration, tool execution, conversation orchestration, and user-facing interfaces, enabling extensibility without architectural entanglement. Orchestral supports advanced agent capabilities found in larger frameworks, including rich tool calling, context compaction, workspace sandboxing, user approval workflows, sub-agents, memory management, and MCP integration.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑