Skip to main content
Back to Docs

Migration Guide

Migrate to AgentiCraft from LangChain, CrewAI, or custom LLM integrations with minimal disruption.

2 min read

Why Migrate

AgentiCraft focuses on infrastructure, not abstractions. You get:

  • Unified provider interface — one API for 14 LLM providers
  • Built-in resilience — circuit breakers, rate limiting, key rotation out of the box
  • Cost optimization — automatic cost-aware routing across providers
  • Type safety — Pydantic v2 models and Protocol-based interfaces
  • No lock-in — standard Python, no custom DSLs or proprietary formats

From Direct API Calls

If you're calling provider SDKs directly:

Before

from openai import AsyncOpenAI
 
client = AsyncOpenAI(api_key="sk-...")
response = await client.chat.completions.create(
    model="gpt-5-mini",
    messages=[{"role": "user", "content": "Hello"}],
)
print(response.choices[0].message.content)

After

from agenticraft_llm import LLMProviderConfig, create_provider
 
config = LLMProviderConfig(provider="openai", model="gpt-5-mini")
provider = create_provider(config)
 
response = await provider.complete(
    messages=[{"role": "user", "content": "Hello"}],
)
print(response.content)

The benefit: swap provider="openai" for provider="anthropic" and your code still works.

From LangChain

LLM Calls

LangChain wraps providers in ChatModel classes. AgentiCraft uses a simpler provider pattern:

# LangChain
from langchain_openai import ChatOpenAI
llm = ChatOpenAI(model="gpt-5-mini")
result = await llm.ainvoke("Hello")
 
# AgentiCraft
from agenticraft_llm import LLMProviderConfig, create_provider
provider = create_provider(LLMProviderConfig(provider="openai", model="gpt-5-mini"))
result = await provider.complete(messages=[{"role": "user", "content": "Hello"}])

Chains

LangChain chains map to sequential function calls. No special abstraction needed:

# LangChain
chain = prompt | llm | parser
result = await chain.ainvoke({"input": "..."})
 
# AgentiCraft — just use async functions
response = await provider.complete(messages=format_prompt(input_data))
result = parse_response(response.content)

From CrewAI

CrewAI's agent/task model maps to AgentiCraft's provider + routing layer:

# CrewAI
from crewai import Agent, Task, Crew
 
agent = Agent(role="Researcher", llm="gpt-5.4")
task = Task(description="Research topic X", agent=agent)
crew = Crew(agents=[agent], tasks=[task])
result = crew.kickoff()
 
# AgentiCraft — focus on the LLM infrastructure
from agenticraft_llm import CostAwareRouter, LLMProviderConfig
 
router = CostAwareRouter(
    providers=[
        LLMProviderConfig(provider="openai", model="gpt-5.4"),
        LLMProviderConfig(provider="anthropic", model="claude-sonnet-4-6"),
    ]
)
# Your agent logic is plain Python — no framework required
result = await router.complete(messages=research_prompt)

Migration Checklist

  • Run uv add agenticraft-llm agenticraft-types
  • Replace direct provider SDK calls with create_provider()
  • Configure key pool for each provider
  • Add circuit breaker and rate limiting
  • Set up cost-aware routing if using multiple providers
  • Remove old provider SDK dependencies (optional — they're still used internally)
  • Update tests to use AgentiCraft's provider interface

Next Steps