Stream‑mode LangChain Agent

What this project does

This repository contains a minimal example that demonstrates how to replace the invoke() call of a LangChain agent with the streaming interface stream(). The output is printed token by token so that the user can see the agent “think” in real time.

The script uses the BroJS LLM provider and a tiny mock tool called get_price to illustrate how tools are called while streaming.

File structure

├── main.py          # Entry point – shows three example invocations
├── requirements.txt # Dependencies (no exact versions)
└── README.md        # This file

Installation

pip install -r requirements.txt
export JOURNAL_MCP_PAT=YOUR_BROJS_TOKEN  # required for the LLM
python main.py

Example output

Running python main.py produces something like:

=== Example 1 ===
--- Step 0 ---
Paris is the capital of France.
--- End ---

=== Example 2 ===
--- Step 0 ---
The price of milk in Kazan is 89 rubles.
--- End ---

=== Example 3 ===
--- Step 0 ---
The price of bread in Kazan is 45 rubles.
The price of coffee in Moscow is 120 rubles.
--- End ---

How it works

• The agent is created with create_agent and a single tool.
• agent.stream() returns an iterator that yields tuples (chunk_type, chunk_data) where chunk_type can be 'messages' or 'updates'.
• Tokens are printed as they arrive, and step changes are highlighted with a separator line.

Feel free to extend the tool set or modify the prompt – the streaming logic remains unchanged.