Mental Support Chat Agent

A LangGraph-based mental support chat agent that interacts with users via the command line. It analyzes user status, detects suicide risk, retrieves knowledge from a local database, and generates supportive responses using a large language model (LLM, default: Gemini Pro 2.5).

Features

• Status Classification – Analyzes the user’s current mental state using a local status_classification model.
• Suicide Risk Detection – Detects potential suicide risk using a local suicide_detection model.
• Knowledge Retrieval – Fetches relevant information from a local knowledge base.
• Response Generation – Sends user input, classification results, and retrieved knowledge to an LLM to generate supportive responses.
• Command-line Chat – Users can interact with the agent directly in the terminal.

Installation & Usage

1. Clone the repository:

git clone <repository_url>
cd <project_directory>

2. Install dependencies:

pip install -r requirements.txt

3. Start the command-line agent:

Copy code
python main.py

Chat with the agent directly in the terminal.

Node Workflow

User Input → status_classification_node → Determine user status

User Input → suicide_detection_node → Detect suicide risk

Knowledge Retrieval → Fetch relevant information

Integrate Information → Send to LLM (gemini_client) → Generate response

Return Response → Display to user

Configuration

LLM Model – Default is Gemini Pro 2.5. Can be changed in gemini_client.py.

Local Models – Paths configurable in status_classifier.py and suicide_detector.py.

Knowledge base

Place knowledges in data\knowledge\jsons

Notes

Ensure local models and the knowledge base are properly deployed.

Suicide risk detection is for reference only. In case of high-risk detection, contact professional help immediately.