Welcome to the Agent Package Manager (APM) documentation. APM is the AI Package Manager that transforms any project into reliable AI-Native Development - build, package, share, and run agentic workflows across any LLM runtime.
New to APM? Follow this learning path for the fastest way to get productive:
- Getting Started - Installation, setup, and your first project
- Core Concepts - Understanding the AI-Native Development framework
- Examples & Use Cases - Real-world workflow patterns and enterprise examples
- Context Guide - Build custom workflows and advanced patterns
- APM Compilation - Mathematical context optimization that solves AI agent scalability
- CLI Reference - Complete command documentation with examples
- Plugin System - Installing and using plugins with plugin.json format
- Integration Guide - VSCode, Spec-kit, AI runtimes, and tool compatibility
- Context Guide - Creating and using the four primitive types
- Prompts Guide - Creating reusable AI instructions and workflows
- Runtime Integration - Setup for Codex and LLM runtimes
- Development Status - Current feature status and testing coverage
- Integration Testing - Testing strategy and execution
- Contributing Guide - How to contribute to the APM ecosystem
- APM Manifesto - Our principles and philosophy for AI-Native Development
- GitHub Issues - Bug reports and feature requests
- AI-Native Development Guide - Complete framework documentation
- MCP Integration - Model Context Protocol tool integration (Phase 2)
APM implements the complete AI-Native Development framework through:
- 🔧 Layer 1: Markdown Prompt Engineering - Structured, repeatable AI instructions
- ⚙️ Layer 2: Context - Configurable tools that deploy prompt + context engineering
- 🎯 Layer 3: Context Engineering - Strategic LLM memory management for reliability
Result: Transform from supervising every AI interaction to architecting systems that delegate complete workflows to AI agents.
Ready to get started? Jump to the Getting Started Guide or explore Core Concepts to understand the framework.