Agent Runtime Overview

关于

This skill serves as a foundational guide for the Agent Runtime monorepo, a general-purpose platform designed to launch and manage AI agent workloads in secure, isolated Modal sandboxes. It outlines the core architecture—connecting React frontend hooks to a Node.js backend—and details essential features like lazy-initialized session management, real-time WebSocket streaming, and support for multiple agent architectures like Claude SDK and OpenCode. Whether you are building custom AI interfaces or multi-tenant automation tools, this skill clarifies how the runtime handles orchestration without enforcing rigid data schemas on the calling application.

主要功能

• Lazy-initialized session management with built-in idle timeout and state syncing
• Multi-architecture support for Claude Agent SDK and OpenCode agents
• Real-time block-by-block streaming of agent output via WebSocket events
• 0 GitHub stars
• Isolated agent execution in Modal sandboxes for secure code environments
• First-class React integration with specialized hooks for messages and workspace tracking

使用场景

• Designing multi-tenant AI applications that require secure, isolated execution environments
• Understanding how to implement custom persistence adapters for agent session storage
• Learning the high-level architecture and data flow of the agent-service monorepo