Developer Guides#
🚧 Early Access Documentation
This documentation is part of an early access release and is under active development. Many sections are still being written, edited, or reorganized. Expect inconsistencies, missing content, outdated references, and broken cross-links.
We welcome feedback! If you find issues or have suggestions, please open an issue on our GitHub page.
Comprehensive learning-oriented guides for building sophisticated agent applications with the Alpha Berkeley Framework. These guides combine architectural understanding with practical implementation patterns, taking you from basic concepts to production-ready systems.
🎯 Learning Paths & Quick Navigation
Essential Learning Path
Understanding the Framework - Core concepts and architecture
Building Your First Capability - Build your first capability
State Management Architecture - Master state patterns
Gateway Architecture - Understand message processing
Human Approval - Deploy with safety
Key Architecture Concepts:
Orchestrator-First Architecture: Upfront Planning in Practice - Why upfront planning works
Convention over Configuration: Configuration-Driven Registry Patterns - Zero-boilerplate development
LangGraph Integration: Native StateGraph and Workflow Execution - StateGraph workflows
Fast Track to Productivity
Building Your First Capability - @capability_node patterns
Message and Execution Flow - Router-controlled architecture
Orchestrator Planning - LLM-powered execution coordination
Python Execution - Container & local execution
Advanced Integration Patterns:
Context Management System - Pydantic context objects
Registry and Discovery - Convention-based loading
Data Integration - Provider framework patterns
Development & Testing
Building Your First Capability - Complete capability implementation
State and Context Essentials - StateManager & ContextManager patterns
Running and Testing - CLI interface & debugging workflows
State & Context Management
State Management Architecture - Selective persistence strategy
Context Management System - Type-safe data exchange
Memory Storage - Cross-session context preservation
Error Handling & Recovery
Error Handling - LLM-powered recovery
Message Generation - Clarification workflows
Security & Approval Systems
Human Approval - LangGraph-native interrupts
Python Execution - Code analysis & container execution
Gateway Architecture - Universal entry point security
Deployment & Integration
Container Deployment - Service orchestration & health monitoring
Data Integration - Provider framework & parallel retrieval
Memory Storage - File-based storage with framework integration
Processing Pipeline
Task Extraction - Conversational context compression
Classification and Routing - Intelligent capability selection
Orchestrator Planning - Complete execution coordination
Implementation Shortcuts
Build a new capability → Building Your First Capability
Add approval workflows → Human Approval
Enable Python execution → Python Execution
Store user context → Context Management System
Deploy with containers → Container Deployment
Handle complex data flows → Message and Execution Flow
Integrate external data → Data Integration
Understand the architecture → Infrastructure Architecture: Classification-Orchestration Pipeline
Debug execution flow → Message and Execution Flow
Customize framework behavior → Convention over Configuration: Configuration-Driven Registry Patterns
Gateway & Message Processing
Gateway Architecture - Universal entry point
Task Extraction - Chat history compression
Message Generation - Adaptive response system
Classification & Orchestration
Classification and Routing - Capability selection logic
Orchestrator Planning - Execution plan creation
Orchestrator-First Architecture: Upfront Planning in Practice - Planning vs reactive patterns
State & Registry Systems
State Management Architecture - AgentState lifecycle
Registry and Discovery - Component discovery & lazy loading
Convention over Configuration: Configuration-Driven Registry Patterns - Decorator-based registration
Guide Categories#
Core concepts, design principles, and the orchestrator-first philosophy that makes the framework powerful and reliable.
Master capability development, state management, and testing workflows. Get productive immediately with convention-based patterns.
State management, context systems, registry patterns, and execution flow. Master the internals for sophisticated applications.
Gateway architecture, task extraction, classification, orchestration, and message generation for intelligent agent behavior.
Human approval workflows, data integration, secure execution, memory storage, and container orchestration for production systems.
📚 Complete Guide Structure
🏛️ Understanding the Framework
Infrastructure Architecture: Classification-Orchestration Pipeline - Gateway, three-pillar pipeline, and system integration
Convention over Configuration: Configuration-Driven Registry Patterns - Configuration-driven loading, registry patterns, and reduced boilerplate
LangGraph Integration: Native StateGraph and Workflow Execution - StateGraph, interrupts, checkpoints, and native features
Orchestrator-First Architecture: Upfront Planning in Practice - Upfront planning vs. iterative tool calling
🚀 Quick Start Patterns
Building Your First Capability - BaseCapability, decorators, and framework integration
State and Context Essentials - AgentState, ContextManager, and data sharing patterns
Running and Testing - Gateway testing, CLI workflows, and debugging techniques
⚙️ Core Framework Systems
State Management Architecture - LangGraph-native state with selective persistence
Context Management System - Type-safe data sharing between capabilities
Registry and Discovery - Component registration and convention-based loading mechanisms
Message and Execution Flow - Complete pipeline from input to response
🔧 Infrastructure Components
Gateway Architecture - Single entry point with state and approval integration
Task Extraction - Natural language to structured task conversion
Classification and Routing - Capability selection and intelligent flow control
Orchestrator Planning - LLM-powered execution plan creation
Message Generation - Adaptive response formatting and clarification
Error Handling - AI-powered error recovery and communication
🏭 Production Systems
Human Approval - LangGraph-native approval with rich context
Data Integration - Parallel retrieval and intelligent provider discovery
Python Execution - Secure code execution with security analysis
Memory Storage - Persistent user memory with framework integration
Container Deployment - Service orchestration and template-based deployment