What is AgentForge?
AgentForge is a production-ready AI agent framework built on top of LangGraph and LangChain. It provides a comprehensive toolkit for building, testing, and deploying AI agents with TypeScript.
Philosophy
AgentForge follows a simple philosophy: We wrap, don't replace. Instead of creating yet another agent framework from scratch, we build on the solid foundation of LangGraph and LangChain, providing:
- Type-safe wrappers around LangGraph primitives
- Production-ready patterns that work out of the box
- Developer experience tools to accelerate development
- Battle-tested utilities for common tasks
Key Features
🛠️ Tool Registry & Auto-Prompt Generation
AgentForge's killer feature - a centralized tool registry that automatically generates LLM-ready prompts:
- Register Once, Use Everywhere - Define tools with rich metadata (categories, tags, examples)
- Auto-Generate Prompts - Automatically create formatted tool descriptions for LLMs
- Smart Querying - Search by name, category, or tags
- Seamless Integration - Convert to LangChain tools with one method call
- 69+ Standard Tools - Production-ready tools included out of the box
const registry = new ToolRegistry();
// Mix custom tools with 69+ standard tools
import { calculator, httpGet, fileReader, webSearch } from '@agentforge/tools';
registry.registerMany([myTool, calculator, httpGet, fileReader, webSearch]);
// Automatically generate LLM-ready prompts
const prompt = registry.generatePrompt({
includeExamples: true,
groupByCategory: true
});
// Use with any agent
const agent = createReActAgent({
model: llm,
tools: registry.toLangChainTools(),
systemPrompt: prompt // Auto-generated!
});
// Create specialized registries for different agents
const webScraperRegistry = new ToolRegistry();
const dataAnalystRegistry = new ToolRegistry();
// Each agent gets its own focused toolset!🎯 Agent Patterns
Four production-ready patterns:
- ReAct - Reasoning and acting in cycles
- Plan-Execute - Planning then executing steps
- Reflection - Self-critique and improvement
- Multi-Agent - Coordinated agent systems
🔌 Middleware System
Composable middleware for:
- Caching and rate limiting
- Validation and error handling
- Logging and tracing
- Metrics and monitoring
- Retry and timeout logic
📊 Streaming & Real-time
- Server-Sent Events (SSE)
- WebSocket support
- Progress tracking
- Backpressure management
🧪 Testing Utilities
- Mock LLM and tools
- Test helpers and assertions
- Sample conversations and fixtures
- Integration testing tools
🚀 Production Features
- Health checks and monitoring
- Resource management
- Connection pooling
- Circuit breakers
- Deployment templates
Architecture
AgentForge is organized into five packages:
@agentforge/core
The foundation package containing:
- Tool Registry - Centralized tool management with auto-prompt generation
- Tool Builder - Fluent API for creating tools with rich metadata
- LangGraph utilities
- Middleware system
- Streaming utilities
- Resource management
- Monitoring tools
@agentforge/patterns
Pre-built agent patterns:
- ReAct pattern
- Plan-Execute pattern
- Reflection pattern
- Multi-Agent coordination
@agentforge/cli
Command-line tool for:
- Project scaffolding
- Development server
- Code generation
- Deployment
@agentforge/testing
Testing utilities:
- Mock factories
- Test helpers
- Fixtures
- Integration testing
@agentforge/tools
Standard tools library:
- Web tools (HTTP, scraping, parsing)
- Data tools (JSON, CSV, XML)
- File tools (read, write, search)
- Utility tools (date, string, math, validation)
When to Use AgentForge
AgentForge is ideal when you need:
✅ Production-ready agents - Not just prototypes
✅ Type safety - Full TypeScript support
✅ Proven patterns - Battle-tested architectures
✅ Smart tool management - Auto-generate prompts, no manual prompt engineering
✅ Developer experience - CLI, testing, docs
✅ LangChain ecosystem - Leverage existing tools
When NOT to Use AgentForge
AgentForge might not be the best fit if:
❌ You need a framework-agnostic solution
❌ You're building simple chatbots (use LangChain directly)
❌ You want to avoid the LangChain ecosystem
❌ You need Python (AgentForge is TypeScript-only)
Note: AgentForge is built on top of LangGraph, not instead of it. You have full access to LangGraph's flexibility and can create custom patterns using AgentForge's utilities while still leveraging all of LangGraph's capabilities.
Comparison with Alternatives
vs. LangChain/LangGraph Directly
- ✅ Higher-level abstractions
- ✅ Production-ready patterns
- ✅ Better TypeScript support
- ✅ Auto-prompt generation - No manual tool description writing
- ✅ More comprehensive tooling
- ✅ Full LangGraph access - Build custom patterns when needed
- ⚠️ Additional abstraction layer (minimal overhead)
vs. AutoGPT/BabyAGI
- ✅ More structured and maintainable
- ✅ Better production features
- ✅ Type safety
- ❌ Less autonomous
vs. CrewAI
- ✅ TypeScript instead of Python
- ✅ Built on LangGraph
- ✅ More comprehensive middleware
- ❌ Smaller community (newer)
Next Steps
Ready to get started? Check out: