Provider Overview

Compare LangChain, CrewAI, AutoGen, Semantic Kernel, Haystack, DSPy.

---

Quick Comparison

Framework	Focus	Best For	Complexity
LangChain/LangGraph	Chains & graphs	Complex workflows	High
CrewAI	Multi-agent teams	Role-based collaboration	Medium
AutoGen (Microsoft)	Conversational	Agent chat/debate	Medium
Semantic Kernel	Enterprise	.NET/enterprise integration	High
Haystack	Pipelines	RAG & search	Medium
DSPy (Stanford)	Programmatic	Prompt optimization	Medium-High
SPINE	Context engineering	Claude Code, MCP-native	Low-Medium

---

Feature Matrix

Feature	LangChain	CrewAI	AutoGen	SPINE
Multi-Provider	✅	⚠️ Limited	✅	✅ Optimized
Self-Play/Debate	⚠️ Manual	⚠️ Critic	✅ Group chat	✅ DIALECTIC
Oscillation Detection	❌	❌	❌	✅
Context Stacks	⚠️ Templates	❌	❌	✅ 6-layer
MCP Native	❌	❌	❌	✅
Cross-Session Memory	⚠️ External	❌	⚠️ External	✅ Minna

---

LangChain / LangGraph

Type: Chain-based orchestration with graph support

from langchain import LLMChain, PromptTemplate

chain = LLMChain(
    llm=ChatOpenAI(),
    prompt=PromptTemplate.from_template("Answer: {question}")
)
result = chain.run("What is 2+2?")

Pros	Cons
Huge ecosystem	Complexity can explode
Many integrations	Rapid API changes
Good documentation	Abstraction overhead

---

CrewAI

Type: Multi-agent team orchestration

from crewai import Agent, Task, Crew

researcher = Agent(role="Researcher", goal="Research thoroughly")
writer = Agent(role="Writer", goal="Write clearly")

crew = Crew(
    agents=[researcher, writer],
    tasks=[research_task, write_task],
    process="sequential"
)
result = crew.kickoff()

Pros	Cons
Intuitive role model	Less flexible
Easy multi-agent	Limited state management

---

AutoGen (Microsoft)

Type: Conversational multi-agent

from autogen import AssistantAgent, UserProxyAgent

assistant = AssistantAgent(name="assistant", llm_config={...})
user_proxy = UserProxyAgent(name="user_proxy")

user_proxy.initiate_chat(assistant, message="Write fibonacci")

Pros	Cons
Natural conversation model	Can be verbose
Built-in code execution	Complex for simple tasks

---

Decision Tree

Do you need multi-agent?
├─ No → Do you need RAG?
│       ├─ Yes → Haystack
│       └─ No → Simple loop (8me Tier 0-1)
│
└─ Yes → Do agents need to debate?
         ├─ Yes → AutoGen or SPINE
         └─ No → Do you need role-based teams?
                 ├─ Yes → CrewAI
                 └─ No → LangGraph

---

The 8me Philosophy

Before reaching for a framework, consider:

1. Do you need it? Simple loops solve many problems
2. Complexity cost: Frameworks add abstraction overhead
3. Debugging difficulty: More layers = harder debugging
4. Lock-in risk: Framework APIs change frequently

---

Next Steps

• Learn about SPINE Integration
• Explore Minna Memory

---

← Back to Concepts