ZHangZHengEric/Sage

What it solves

Sage provides a unified platform for building, deploying, and managing AI agents that can handle complex project work and tool execution. It bridges the gap between defining an agent's capabilities (instructions and tools) and executing them in a reliable, stateful way across different environments, whether on a desktop, a web server, or embedded within a custom Python application.

How it works

Sage is built around a core runtime called SAgents v2, which manages the lifecycle of agents, sessions, and runs. It uses a plugin-based architecture to compose models, memory, storage, and scheduling providers. Agents are defined using declarative sage.yaml packages that specify their instructions and capabilities. The system supports stateful execution with durable session history and the ability to pause and resume tasks for human approval. It also integrates with the Model Context Protocol (MCP) and a reusable skill ecosystem to extend agent functionality.

Who it’s for

• Developers who want to embed agentic capabilities into their own Python applications.
• Users who want a ready-to-use desktop or web-based interface for managing AI agents.
• Architects looking for a modular framework to orchestrate multi-agent collaboration and complex workflows.

Highlights

• Declarative Configuration: Define agent behavior and runtime settings in YAML files.
• Multi-Host Support: Deploy via a dedicated desktop app, a multi-user server, or as a Python library.
• Stateful Interaction: Supports streamed events and human-in-the-loop approvals to ensure reliable delivery.
• Extensible Tooling: Integrates built-in tools, custom skills, and MCP services.
• Multi-Agent Orchestration: Coordinate multiple agents through directed messaging or structured Agent Flows.