AgenticFORGE
A monorepo TypeScript framework for building tool-driven AI agents, providing LLM abstractions, multiple agent paradigms, a skill system, built-in memory and RAG, and inter-agent communication protocols.
AgenticFORGE is a monorepo TypeScript framework for building agents around a standardized Tool, ToolRegistry and ToolChain model with an async execution pipeline. It supports several agent paradigms in one framework: Simple, FunctionCall, ReAct, PlanSolve, Reflection, SkillAgent and WorkflowAgent.
Its skill system lets skills be defined as SKILL.md files or TypeScript classes, with a dispatcher that routes to the right skill first by keyword at zero LLM cost and falls back to LLM-based intent matching. Built-in memory covers working, episodic, semantic and perceptual memory with pluggable stores and a retrieval-augmented generation pipeline, and a protocol layer implements MCP, A2A and ANP for communication between agents.
The framework is split into packages such as core, tools, agents, workflow, skills, memory and protocols, published as ESM and CJS builds with strict TypeScript typing, and a single kit package re-exports the whole ecosystem for one-stop use.