LangChain LCEL building blocks and interfaces

 1. Building Blocks (Modular Components)

These are the functional units used to build chains and agents.

Title	Description
Prompts	Templates that translate user input into a format the model understands.
Models	LLMs or Chat Models that process input and generate language responses.
Output Parsers	Tools that convert raw model output into structured formats (JSON, lists, strings).
Retrievers	Logic for fetching relevant documents from data sources to provide context.
Tools	External functions (like Google Search or calculators) that an agent can call.
Vector Stores	Databases designed to store and query text embeddings for semantic search.

2. Core Interfaces

These define the standards for how data, history, and feedback are handled across the framework.

Title	Description
Runnable	The primary interface for LCEL, providing standard methods like invoke and stream.  (invoke, ainvoke - asynchronous invoke, stream)
BaseChatMessageHistory	Standardizes how conversation messages are saved and retrieved from memory.
BaseStore	A universal interface for simple key-value storage used in caching and persistent state.
BaseCallbackHandler	An interface for building custom listeners to log, monitor, or stream internal events.
BaseRetriever	A specialized search interface that defines how documents are queried.
BaseLanguageModel	The root interface defining how all LLMs and Chat Models must behave.

Runnable methods :  invoke, ainvoke and stream

• invoke: Executes the component synchronously with a single input.
  • Model: Returns the complete generated message.
  • Agent: Executes the entire workflow (including tool calls) and returns the final response.
• ainvoke: The asynchronous version of invoke, recommended for production web backends to prevent blocking the event loop.
• stream: Incremental output generation.
  • Model: Streams the response token-by-token as they are generated.
  • Agent: Streams the agent's "steps" or progress. By setting stream_mode="updates", you can receive real-time updates after each tool call or reasoning step. 

Key Differences in Streaming

While both support these methods, the output they produce during streaming differs:

• Models primarily stream text tokens.
• Agents (especially those built with LangGraph) can stream a wider variety of data, including model reasoning, tool call arguments, tool results, and the final answer.