Phinite 多智能体操作系统：提供一级智能体身份、可组合技能与行为评估
英文摘要
Phinite launched a multi-agent operating system that provides a registry for first-class agent identity (ID, version, owner, skill graph). It replaces traditional unit tests with behavioral evaluation, using compound reliability scoring and behavioral regression to handle non-deterministic agent execution. Skills are versioned, reusable, and agent-inheritable, enabling composability without rebuilding. The platform is cloud-agnostic, model-agnostic, and includes built-in observability (traces, cost attribution, drift detection). It is SOC 2 Type II compliant and offers free credits for testing.
中文摘要
Phinite 发布了一个多智能体操作系统，提供一级智能体身份注册表（包括 ID、版本、所有者、技能图谱），并采用复合可靠性评分与行为回归来取代传统单元测试，以应对智能体的非确定性行为。其技能支持版本化、可复用和智能体继承，无需重构即可实现组合。平台不依赖特定云或模型，内置可观测性（追踪、成本归因、漂移检测），且已通过 SOC 2 Type II 合规认证，提供免费测试积分。
关键要点
First-class agent identity: every agent gets a unique ID, version, owner, and a skill graph in a central registry, analogous to service mesh IAM.
一级智能体身份：每个智能体在中央注册表中拥有唯一 ID、版本、所有者及技能图谱，类似服务网格的 IAM。
Behavioral evaluation instead of unit tests: because agents are non-deterministic, the system uses compound reliability scoring and behavioral regression to assess performance.
行为评估替代单元测试：由于智能体具有非确定性，系统采用复合可靠性评分与行为回归来评估表现。
Composable, reusable skills: skills are versioned and inheritable across agents, inspired by Kubernetes operators, eliminating the need to rebuild when adding capabilities.
可组合、可复用的技能：技能版本化且可跨智能体继承，灵感来自 Kubernetes 操作员模式，增加能力时无需重新构建智能体。
Cloud-agnostic, model-agnostic, with built-in observability: supports any deployment environment and AI model, includes tracing, cost attribution, and drift detection.
云与模型无关，内置可观测性：支持任意部署环境和 AI 模型，提供追踪、成本归因和漂移检测功能。
SOC 2 Type II compliant and offering free credits for testing this week.
已通过 SOC 2 Type II 合规认证，本周提供免费积分供测试。