LangChain Config Generator

LCEL is the modern way to compose LangChain components using the pipe (|) operator. A basic chain looks like: chain = prompt | llm | StrOutputParser(). LCEL chains are composable, support streaming out of the box, and work with .invoke(), .stream(), .batch(), and async variants. LangChain recommends LCEL over legacy LLMChain for all new projects.

SequentialChain is the legacy approach where each chain's output becomes the next chain's input. LCEL replaces this with cleaner pipe composition: chain = prompt1 | llm | prompt2 | llm. Use LCEL for new projects — it's more explicit, supports parallel execution (RunnableParallel), and has better streaming support. SequentialChain is kept here for compatibility with older codebases.

In LangChain PromptTemplate, any word wrapped in curly braces like {name} or {question} becomes a variable. At runtime, call chain.invoke({'name': 'Alice', 'question': 'What is LangChain?'}) to inject values. ChatPromptTemplate.from_template() works the same way for LCEL chains. Variables are automatically detected and listed in this tool.

PromptTemplate produces a single string for completion-style models. ChatPromptTemplate produces a list of messages with roles (system, human, ai) for chat models. Since all modern models (GPT-4o, Claude, Gemini) are chat models, ChatPromptTemplate is preferred. LCEL chains use ChatPromptTemplate.from_template() by default.

For LCEL chains, use RunnableWithMessageHistory to add conversation memory: chain_with_history = RunnableWithMessageHistory(chain, get_session_history, input_messages_key='input', history_messages_key='history'). Pass session_id in config when invoking. For persistent storage, use Redis or SQLite-backed message history classes from langchain_community.

For general-purpose chat: GPT-4o or Claude 3.5 Sonnet offer the best quality. For cost-efficient high-volume: GPT-4o mini or Gemini 1.5 Flash. For latency-sensitive apps: GPT-3.5 Turbo or GPT-4o mini. Always test your specific task — model rankings vary significantly by use case. Use LangSmith to trace and evaluate chain performance.