migrating-to-lauren-ai

name	migrating-to-lauren-ai
description	Ports AI agent code from LangChain or the raw OpenAI SDK to the lauren-ai framework. Provides side-by-side equivalents for chains, agents, tools, memory, and prompts. Use when converting LangChain or OpenAI-SDK-based agent code to lauren-ai, or when a user is familiar with those frameworks and needs the lauren-ai equivalents.

Migrating to Lauren AI

Use codemap find "SymbolName" to locate any symbol before reading — it gives exact file + line range and is faster than grep.

Quick comparison table

LangChain	OpenAI SDK	Lauren AI
Tool / @tool	function + tools=[{...}]	@tool() async function or class
AgentExecutor	manual loop + tool dispatch	AgentRunnerBase / AgentRunner[X]
ChatOpenAI / ChatAnthropic	openai.Client / anthropic.Anthropic	LLMModule.for_root(LLMConfig(...))
ConversationBufferMemory	manual message list	InMemoryConversationStore
ChatPromptTemplate	[{"role": ..., "content": ...}] list	@agent(system=...) + message list
RunnableSequence / `	` chain	manual loop
BaseTool.invoke	tool_call dispatch	ToolContext injected by the runner
AgentType.OPENAI_FUNCTIONS	assistants API	@agent + @use_tools(...)
LLMChain	single chat.completions.create	LLMService.complete() (low-level)

• LangChain migration: from-langchain.md
• OpenAI SDK migration: from-openai-sdk.md

The key conceptual shift

LangChain and the OpenAI SDK are call-time frameworks: you assemble chains or message lists and call them. Lauren AI is declaration-time: you declare agents and tools with decorators, and the framework builds the agentic loop, tool dispatch, memory management, and streaming for you.

# OpenAI SDK — manual loop
while True:
    response = client.chat.completions.create(model="gpt-4o", messages=messages, tools=tools)
    if response.choices[0].finish_reason == "tool_calls":
        # dispatch each tool call manually
        ...
    else:
        break

# Lauren AI — declarative loop
@agent(model="gpt-4o", system="...", max_turns=10)
@use_tools(MyTool, AnotherTool)
class MyAgent: ...

result = await runner.run(my_agent, "user message")  # loop runs automatically