littleblacky/agenticforge

特工锻造

构建具备工具、内存、技能和协议的生产准备型 TypeScript AI 代理。

中文 |英语

---

概述

AgenticFORGE 是一个用于构建工具驱动 AI 代理的单代码仓库 TypeScript 框架。 它提供了从核心LLM抽象到高级代理工作流、技能路由、内存+RAG、内置工具和多代理通信协议的完整栈。

如果你想要一个统一的 SDK，从简单的聊天机器人构建成生产多代理系统，可以从 @agenticforge/kit 开始。

---

为什么是AgenticFORGE

• 工具优先架构：标准化的Tool、ToolRegistry、ToolChain和异步执行模型
• 多代理范式：Simple、FunctionCall、ReAct、PlanSolve、Reflection、SkillAgent和WorkflowAgent
• 技能系统：定义带有SKILL.md或TypeScript类的技能;SkillDispatcher按关键词自动路由（LLM成本为零），然后是LLM意图;withSkills mixin为任意代理类型添加技能路由
• 内置内存 + RAG 内存：工作/情节/语义/感知记忆，带可插拔存储

AgenticFORGE

Build production-ready TypeScript AI agents with tools, memory, skills, and protocols.

中文 | English

---

Overview

AgenticFORGE is a monorepo TypeScript framework for building tool-driven AI agents. It provides a complete stack from core LLM abstractions to advanced agent workflows, skill routing, memory + RAG, built-in tools, and multi-agent communication protocols.

If you want one unified SDK to build from a simple chatbot to a production multi-agent system, start with @agenticforge/kit.

---

Why AgenticFORGE

• Tool-first architecture: Standardized Tool, ToolRegistry, ToolChain, and async execution model
• Multiple agent paradigms: Simple, FunctionCall, ReAct, PlanSolve, Reflection, SkillAgent, and WorkflowAgent
• Skill system: Define skills with SKILL.md or TypeScript classes; SkillDispatcher auto-routes by keyword (zero LLM cost) then LLM intent; withSkills mixin adds skill routing to any Agent type
• Memory + RAG built-in: Working / episodic / semantic / perceptual memory with pluggable stores
• Protocol layer included: MCP, A2A, and ANP implementations for inter-agent communication
• Production-friendly TypeScript: ESM/CJS builds, strict typing, modular packages, and subpath imports

---

Package Architecture