Add solution for streaming tokens from LangGraph nodes (fixes #1539)

src/oss/langgraph/streaming-tokens-example.md

@@ -0,0 +1,126 @@
+# Streaming Tokens from LangGraph Nodes
+
+## Issue Reference
+This document addresses [Issue #1539](https://github.com/langchain-ai/docs/issues/1539) - How to stream individual tokens from LLM calls within LangGraph nodes.
+
+## Problem
+When using `graph.astream()`, you can stream the aggregated output of each node, but not the raw token-level stream produced during the model invocation. The user wants to stream individual tokens as they are generated by the LLM.
+
+## Solution
+
+There are two main approaches to stream individual tokens from LangGraph:
+
+### Approach 1: Use `stream_mode="messages"`
+
+The `messages` streaming mode allows you to stream LLM tokens as they are generated:
+
+```python
+from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
+from langgraph.graph import StateGraph
+
+# Your graph setup here
+graph = ... # your compiled graph
+
+# Stream with messages mode to get token-by-token output
+async for msg, metadata in graph.astream(
+    {"messages": [HumanMessage(content="Your query here")]},
+    stream_mode="messages",
+):
+    # msg.content contains the individual token
+    # metadata contains information about which node generated it
+    if msg.content:
+        print(msg.content, end="", flush=True)
+```
+
+### Approach 2: Use `astream_events` for Fine-Grained Control
+
+For more detailed control, including filtering by specific nodes or tags, use `astream_events`:
+
+```python
+from langchain_core.runnables import RunnableConfig
+
+async def fundamentals_analyst_node(state, config: RunnableConfig):
+    """
+    Important: Add 'config: RunnableConfig' parameter to your node function
+    and pass it to the model invocation for streaming to work.
+    """
+    prompt = ChatPromptTemplate.from_messages([...])
+    chain = prompt | llm.bind_tools(tools)
+
+    # Pass config to enable streaming
+    result = await chain.ainvoke(state["messages"], config)
+
+    return {"messages": [result]}
+
+# Stream events from the graph
+async for event in graph.astream_events(
+    {"messages": [HumanMessage(content="Your query")]},
+    version="v2"
+):
+    # Filter for token chunks
+    if event["event"] == "on_chat_model_stream":
+        chunk = event["data"]["chunk"]
+        if hasattr(chunk, "content") and chunk.content:
+            print(chunk.content, end="", flush=True)
+```
+
+### Key Points
+
+1. **Add `config` parameter**: Your node functions must accept a `config: RunnableConfig` parameter and pass it to model invocations
+2. **Use `stream_mode="messages"`**: Simplest way to get token-by-token streaming
+3. **Use `astream_events`**: For more control over which events to stream
+4. **Metadata filtering**: Use metadata fields like `langgraph_node` or custom tags to filter streams from specific nodes
+
+### Complete Example
+
+```python
+from typing import Annotated
+from typing_extensions import TypedDict
+from langchain_core.messages import AIMessage, HumanMessage
+from langchain_core.prompts import ChatPromptTemplate
+from langchain_core.runnables import RunnableConfig
+from langchain_openai import ChatOpenAI
+from langgraph.graph import StateGraph, END
+from langgraph.graph.message import add_messages
+
+class State(TypedDict):
+    messages: Annotated[list, add_messages]
+
+model = ChatOpenAI(model="gpt-4", streaming=True)  # Enable streaming
+
+# Important: Accept config parameter
+async def my_node(state: State, config: RunnableConfig) -> State:
+    prompt = ChatPromptTemplate.from_messages([
+        ("system", "You are a helpful assistant."),
+        ("placeholder", "{messages}")
+    ])
+    chain = prompt | model
+
+    # Pass config to enable streaming
+    response = await chain.ainvoke({"messages": state["messages"]}, config)
+    return {"messages": [response]}
+
+builder = StateGraph(State)
+builder.add_node("assistant", my_node)
+builder.set_entry_point("assistant")
+builder.add_edge("assistant", END)
+graph = builder.compile()
+
+# Stream tokens
+async for msg, metadata in graph.astream(
+    {"messages": [HumanMessage(content="Tell me a joke")]},
+    stream_mode="messages"
+):
+    if msg.content and metadata.get("langgraph_node") == "assistant":
+        print(msg.content, end="|", flush=True)
+```
+
+## Additional Resources
+
+- [LangGraph Streaming Documentation](https://docs.langchain.com/oss/python/langgraph/streaming)
+- [LangChain Streaming Guide](https://docs.langchain.com/oss/python/langchain/streaming)
+
+## Related Issues
+
+- [GitHub Issue #137](https://github.com/langchain-ai/langgraph/issues/137)
+- [GitHub Discussion #533](https://github.com/langchain-ai/langgraph/discussions/533)