Agent Server

Overview

Agent Server is a FastAPI-based application designed to interact with language models using langchain and langgraph libraries. It provides an API for generating Python code solutions based on user queries and includes a workflow that checks and reflects on the generated code's correctness. The server integrates langserve for easy deployment of the language model and workflow.

Features

• Chat-based Code Generation: Uses a predefined prompt to generate Python code based on user queries.
• Streaming API: Real-time streaming responses for interactive chat applications.
• Deploying Agents: LangServe offers a playground for users to make queries and obtain generated code.
• Code Execution and Validation Funtions: Generates, executes, and validates code function calling.
• Reflective Workflow Function: Analyzes and provides reflections on code generation errors.
• Cross-Origin Resource Sharing (CORS): Enabled for flexibility in accessing the API.

Prerequisites

• Python 3.8+
• FastAPI
• Uvicorn
• langchain
• langserve
• OpenAI API Key

Installation

1. Clone the repository:

   git clone <https://github.com/Workforce-41/production-manager-agent/tree/main>
   cd <production-manager-agent>

2. **Create a Clean Virtual Environment:**

    python -m venv myenv

3. **Install dependencies:**

    pip install -r requirements.txt

4. Set the environment variable in a .env file:

    OPENAI_API_KEY = <YOUR API KEY>

## API Endpoints

1. **/openai**

Description: Endpoint for interacting with the language model.
Method: POST

2. **/chain**
Description: Endpoint for code generation using a LangChain model.
Method: POST

3. **/stream_chat/**
Description: Streaming endpoint for real-time chat responses.
Method: POST
Request Body: JSON object containing messages (string)

4. **/agents**
Description: Workflow endpoint for generating, validating, and reflecting on code solutions.
Method: POST

**Example Request**
To send a request to the /stream_chat/ endpoint:

    json
    {
    "messages": "How can I implement a bubble sort algorithm in Python?"
    }

**Agent Server**:
Follow the link after running app.py: http://localhost:8000/docs
![alt text](image.png)

**Deployed Agent**:
Follow the link after running app.py: http://localhost:8000/agents/playground/
![alt text](image-1.png)

Built With

• fastapi
• langchain
• langserve
• openai
• python