vamplabAI/sgr-agent-core

What it solves

SGR Agent Core provides a framework for building intelligent research agents that can perform complex reasoning and tool selection. It addresses the need for structured, reliable reasoning processes in AI agents, moving beyond simple prompt-based tool calling to a more systematic approach called Schema-Guided Reasoning (SGR).

How it works

The framework implements a two-phase architecture based on the SGR concept. It provides a BaseAgent interface that developers can extend to create specialized agents. The system supports multiple agent types, including SGRAgent, ToolCallingAgent, and SGRToolCallingAgent, and integrates with any OpenAI-compatible LLM (including local models).

To facilitate research, it includes built-in tools for web search, page content extraction, and reasoning. It can be deployed as a REST API server (OpenAI-compatible) or used as a Python library, and it also supports the Agent Client Protocol (ACP) for communication via stdio.

Who it’s for

• AI Developers looking to build research-oriented agents with structured reasoning.
• Software Engineers wanting to integrate agentic workflows into existing applications via a REST API.
• Researchers who need private, local LLM-powered research tools.

Highlights

• Schema-Guided Reasoning: Combines structured reasoning with flexible tool selection.
• OpenAI Compatibility: Features a drop-in replacement for OpenAI API endpoints and works with any OpenAI-compatible LLM.
• Real-time Streaming: Built-in support for streaming responses via Server-Sent Events (SSE).
• Extensible Architecture: Easy to create custom agents and tools using the provided BaseAgent interface.
• Agent Client Protocol Support: Enables integration with editors and tools via JSON-RPC over stdio.