Stream‑Mode AI Agent

A lightweight LangChain agent that prints its output as it is generated using the .stream() API. The project demonstrates how to replace a single .invoke() call with streaming so that users see partial responses in real time.

Why stream?
When an LLM produces long answers or calls multiple tools, waiting for the whole result can feel like a freeze. Streaming gives instant feedback and improves UX.

📦 Project Structure

.
├── agent.py   # Agent definition with streaming support
├── client.py  # Simple CLI to interact with the agent
└── README.md

agent.py – creates an agent that uses a single tool (get_price) and streams its output.
client.py – runs the agent in a loop, reading user queries from stdin.

⚙️ Prerequisites

Item	Version
Python	3.10+
pip	latest
LangChain	`>=0.2.0`
Rich	`>=13.0`

Tip: Use a virtual environment to keep dependencies isolated.

🚀 Installation

# Clone the repo (or copy the files)
git clone https://github.com/your-username/stream-agent.git
cd stream-agent

# Create and activate a venv (optional but recommended)
python -m venv .venv
source .venv/bin/activate   # Windows: .venv\Scripts\activate

# Install dependencies
pip install --upgrade pip
pip install langchain rich

If you plan to use an OpenAI model, set the API key:
export OPENAI_API_KEY="sk-..."

📖 Usage

Run the client script and type your questions. The agent will stream its answer token‑by‑token.

python client.py

Example session:

$ python client.py
Enter your query (Ctrl+C to exit):
> What is the price of Bitcoin today?

Agent: Checking price...
Agent: Current BTC price is $42,300.00

The output appears progressively as the LLM generates it.

🛠️ Customization

Add more tools – decorate a function with @tool and pass it to create_agent.
Change model – modify the model_name argument in agent.py.
Adjust streaming format – tweak the stream() call or use Rich’s live rendering for fancy UI.

2.2 KiB Raw Blame History Unescape Escape