A framework for building, orchestrating, and deploying production-grade AI agents and multi-agent workflows in Go. It targets development teams aiming to move beyond simple chatbots to complex, stateful agent systems. MAF-Go offers a consistent foundation for agent operations, supporting diverse architectural choices and a broad ecosystem including Azure OpenAI, OpenAI, and Microsoft Foundry, enabling flexible and evolving agent architectures.

How It Works

MAF-Go employs a modular design centered around Go packages for agents, tools, middleware, and workflows. Its core strength lies in its flexible middleware system, which handles request/response processing, logging, OpenTelemetry integration, and automatic tool calling. The framework supports advanced orchestration patterns, including sequential, concurrent, and collaborative multi-agent workflows, with features like checkpointing, streaming, and human-in-the-loop control. This approach allows for complex agent interactions and robust operational capabilities.

Quick Start & Requirements

Installation is via go get github.com/microsoft/agent-framework-go. Development requires a Go environment and authentication setup for cloud providers (e.g., az login for Azure credentials). Environment variables are used for configuration (e.g., FOUNDRY_PROJECT_ENDPOINT, FOUNDRY_MODEL). Links to official quick-start guides, tutorials, and detailed examples for agents, workflows, and specific providers are available within the repository.

Highlighted Details

• Comprehensive Go SDK for agent and workflow development.
• Support for multiple LLM and agent providers (e.g., Azure OpenAI, OpenAI, Foundry).
• Extensible middleware for request/response handling, observability (OpenTelemetry), and tool integration.
• Advanced orchestration patterns: sequential, concurrent, group collaboration, conditional routing, subworkflows, checkpointing, streaming, human-in-the-loop, and time-travel.
• Integration with Microsoft Foundry for project-backed agents.
• Agent Skills for incorporating domain-specific knowledge.

Maintenance & Community

The project is actively maintained by Microsoft and is in public preview. Community support and bug reporting are primarily handled through GitHub issues. A Discord server is available for discussions, though a specific invite URL is not provided in the README.

Licensing & Compatibility

The specific license is not detailed in the provided README. Users should exercise caution when integrating with third-party systems, as these are governed by their own terms and licenses.

Limitations & Caveats

MAF-Go is in public preview, with some features still under development or not yet implemented, such as declarative agents, handoff orchestration, AF Labs, and DevUI. The .NET implementation currently offers broader product integrations. Users are responsible for managing data flow and responsible AI practices when integrating with third-party systems.