Back to articles
AI & Cloud

Deploy Multi-Task AI Coding Agent on 4GB RAM VPS with Qwen Agent and LiteLLM

A detailed guide to deploying a multi-task AI coding agent on a 4GB RAM VPS using Qwen Agent and LiteLLM, offering a self-hosted alternative to Cursor Bolt New. This solution...

DPTCloud Engineering4 min read
Deploy Multi-Task AI Coding Agent on 4GB RAM VPS with Qwen Agent and LiteLLM

A detailed guide to deploying a multi-task AI coding agent on a 4GB RAM VPS using Qwen Agent and LiteLLM, offering a self-hosted alternative to Cursor Bolt New. This solution enables developers to automate coding tasks, integrate multiple AI models, and manage inference via a unified API gateway—all without relying on cloud services. The setup runs efficiently on modest hardware, leveraging Docker and lightweight models for optimal performance.

Deploying a Multi-Task AI Coding Agent on a 4GB RAM VPS

As demand for AI-assisted programming grows, self-hosting an AI coding agent offers developers greater control, privacy, and cost efficiency. With a 4GB RAM VPS, you can run a capable multi-task AI coding agent using Qwen Agent and LiteLLM—two open-source tools that support model aggregation, automation, and local execution. This guide walks through deploying a fully self-hosted system that replaces cloud-dependent tools like Cursor Bolt New, enabling local AI coding assistance with minimal infrastructure overhead.

Overview of the Solution: Qwen Agent + LiteLLM

Qwen Agent, developed by Alibaba Cloud, is an AI agent capable of writing, debugging, analyzing, and documenting code. When paired with LiteLLM, a unified API gateway for multiple large language models, you can create a powerful, self-hosted coding assistant. LiteLLM abstracts access to various models—Qwen, GPT, Claude, Mistral, and others—through a single endpoint, simplifying integration and reducing complexity.

On a 4GB RAM VPS, you can run both Qwen Agent and LiteLLM simultaneously without a GPU by using optimized or quantized models. LiteLLM supports model routing, rate limiting, and fallback strategies, making it ideal for production-like environments where reliability and flexibility are key.

Prerequisites and Installation

Prerequisites

  • 4GB RAM or higher VPS (Ubuntu 20.04/22.04 recommended)
  • Linux operating system
  • Docker and Docker Compose
  • Python 3.9 or higher
  • Git

Installation

Run the following commands on your VPS to install required tools:

sudo apt update
sudo apt install -y docker.io docker-compose git python3 python3-pip
sudo usermod -aG docker $USER
newgrp docker

Restart your system or re-login to apply Docker group changes.

Deploying Qwen Agent and LiteLLM

Create a project directory and clone the repositories:

mkdir ai-coding-agent
cd ai-coding-agent
git clone https://github.com/qwen-agent/qwen-agent.git
git clone https://github.com/BerriAI/litellm.git

Now, create a docker-compose.yml file to orchestrate both services:

version: '3.8'

services:
  qwen-agent:
    image: qwen-agent:latest
    container_name: qwen-agent
    ports:
      - "8080:8080"
    volumes:
      - ./data:/data
    environment:
      - QWEN_API_KEY=your_qwen_key
      - MODEL_NAME=qwen-7b
    restart: unless-stopped

  litellm:
    image: berriai/litellm:latest
    container_name: litellm
    ports:
      - "8000:8000"
    environment:
      - LITELLMMODELS=qwen-7b,gpt-3.5-turbo,claude-3-haiku
      - LITELLMSERVER=true
    restart: unless-stopped

Start the services:

docker-compose up -d

Verification and Testing

After deployment, verify the services are running:

docker ps

Check the health endpoints:

  • http://<your-vps-ip>:8080/health – Qwen Agent health check
  • http://<your-vps-ip>:8000/health – LiteLLM health check

Test the API from a client:

curl -X POST http://<your-vps-ip>:8000/completion \
  -H "Content-Type: application/json" \
  -d '{\"model\": \"qwen-7b\", \"prompt\": \"Write a Python function to calculate factorial\"}'

Replacing Cursor Bolt New

Cursor Bolt New is a cloud-based AI coding tool that may raise concerns about data privacy, ongoing costs, and network dependency. By using a self-hosted setup with Qwen Agent and LiteLLM, you gain:

  • Full control over source code and data
  • Lower long-term operational costs
  • Integration with existing CI/CD pipelines and IDEs
  • Customizable AI models based on workload needs

This approach delivers similar functionality to Cursor Bolt New while offering greater flexibility, security, and scalability.

Troubleshooting

  • API connection failed: Check if Docker Compose is running, ports are open, and firewall rules allow traffic.
  • Memory exhaustion: Use smaller models (e.g., Qwen-1.8B), enable model quantization, or upgrade RAM.
  • API returns errors: Verify environment variables, inspect Docker logs, and validate model configuration.

References

DPTCloud Tier III Infrastructure

High-RAM Cloud VPS Plans Optimized for AI, Databases & Cache

Enterprise RAID 10 NVMe SSDs, CMC & FPT District 9 uplinks with Anycast Anti-DDoS and 99.99% SLA.

View 22 VPS Plans
ECO · vn-02Recommended

VPS-4E

297.000 ₫ / mo

10% VAT included

  • CPU: 4 vCPU
  • RAM: 8 GB
  • SSD: 50 GB NVMe
ECO · vn-02

VPS-5E

614.900 ₫ / mo

10% VAT included

  • CPU: 8 vCPU
  • RAM: 16 GB
  • SSD: 100 GB NVMe
ADVANCED · vn-03Recommended

VPS-4T

548.900 ₫ / mo

10% VAT included

  • CPU: 4 vCPU
  • RAM: 8 GB
  • SSD: 50 GB NVMe

Related Articles

View all
ZaloZalo SupportDiscord0384 753 478