Frameworks
AI frameworks provide the abstractions needed to build, optimize, and deploy agentic and RAG-based applications. They handle the "glue" code of LLM interactions, tool execution, and prompt management.
Framework Selection Guidance
| Goal |
Recommended Frameworks |
Why? |
| General Purpose RAG |
LlamaIndex, LangChain |
Mature ecosystems with deep data and tool integrations. |
| Multi-Agent Systems |
AutoGen, CrewAI, AG2 |
Specialized in agent coordination, delegation, and role-playing. |
| Structured Output |
Instructor, PydanticAI |
Focus on typed, reliable data extraction using Pydantic. |
| Optimization |
DSPy |
Programmatic prompt optimization instead of manual trial-and-error. |
| Local / Lightweight |
Smolagents, Mastra |
Minimalist approach with focus on speed and developer experience. |
Core Framework List
| Framework |
Primary Language |
Role |
| AG2 |
Python |
Advanced multi-agent orchestration. |
| AutoGen |
Python |
Original multi-agent conversation framework. |
| CrewAI |
Python |
Role-based agent collaboration. |
| DSPy |
Python |
Prompt compiler and optimizer. |
| Haystack |
Python |
Modular pipeline framework for RAG. |
| LangChain |
Python / JS |
Swiss-army knife for LLM apps. |
| LlamaIndex |
Python / JS |
Context-augmented data framework. |
| Mastra |
TypeScript |
Integration-first agent engine. |
| PydanticAI |
Python |
Typed, functional agent framework. |
| Semantic Kernel |
C# / Python |
Microsoft-native agent SDK. |