Cloudways, a leading managed PHP hosting service, recently introduced Cloudways Copilot, an AI-powered Site Reliability Engineer aimed at revolutionizing their support operations. With a fleet of over 90,000 servers and half a million applications, Cloudways faced a significant challenge in managing the growing support load. By leveraging AI technology, Cloudways aims to streamline their support processes and provide faster, more efficient solutions to their customers.
The AI SRE agent, Cloudways Copilot, offers features such as Insights and SmartFix, which provide users with detailed diagnosis and resolution steps for web app incidents. These AI-powered insights are faster and more consistent than those provided by human agents, ensuring a quicker resolution to issues.
The architecture of Cloudways Copilot includes a monitoring layer that continuously observes user machines for web stack issues and anomalies. When an issue is detected, an alert is triggered and forwarded to the control plane, which routes it to the Insight Generation Engine. This engine consists of three components: the AI SRE Agent, Orchestration Layer, and DigitalOcean serverless inference.
The AI SRE Agent is optimized to work with Cloudways’ customizations on top of Debian, providing details such as file structure, log file locations, system navigation commands, and core services needed for web apps to run. The agent is hosted on the DigitalOcean Kubernetes platform.
The Orchestration Layer connects the AI Agent with the fleet of servers using Ansible Server. When the AI Agent initiates an SSH command, it is stored on Redis and executed sequentially through a Celery queue. This ensures data security on user machines and efficient communication between the AI Agent and the servers.
DigitalOcean serverless inference plays a critical role in gathering information and generating insights for users. This component is easy to set up and involves invoking a single API endpoint for inference.
After generating insights, Cloudways implements a post-generation insight review process to validate the accuracy and quality of the information provided. This involves both manual reviews by human agents and a secondary evaluation layer where another AI Agent reviews outputs for correctness.
Identifying the right problems to solve with AI is crucial for maximizing its benefits. Cloudways focuses on tasks that are repetitive, time-consuming for humans, and operationally critical, such as finding directories to recover disk space, tracing bots generating excessive requests, identifying problematic configuration rules causing downtime, and detecting resource-hogging processes.
Fine-tuning AI models for common SRE and infrastructure problems is often unnecessary, as modern state-of-the-art models already have a deep understanding of these domains. Cloudways aims to leverage AI technology to enhance their support operations and provide faster, more efficient solutions to their customers. When it comes to utilizing AI agents effectively, it’s crucial to use the right tool for the right job. An AI agent is a program where Large Language Models (LLMs) output control parts of the workflow, but it’s important to remember that not every step should be handled by the model itself. Deterministic tasks should be implemented in code, while AI should be utilized for reasoning and pattern recognition.
One key aspect to consider when working with LLMs is to accept non-determinism. LLMs are inherently non-deterministic, and trying to force deterministic behavior from them can lead to frustration and poor system design. Instead, it’s advisable to build systems that can tolerate variability and validate outputs effectively.
While AI agents have the ability to replicate human operators’ tasks at scale, it’s essential not to overestimate their capabilities. AI is not a magical solution and relies heavily on structured knowledge and historical context. In environments that are too novel, poorly documented, or inconsistent, AI performance may suffer.
It’s also important to avoid the sunk cost trap when working with AI agents. The field of AI-based tooling is still evolving, and it’s possible to discover that LLMs are not the right solution for a particular problem. In such situations, it’s important to be pragmatic and focus on what delivers value while abandoning what doesn’t.
When running AI agents at scale, having a reliable and flexible inference partner is crucial. The DigitalOcean Gradient AI Platform is designed to meet these needs effectively. The platform offers features such as Serverless inference, support for both open-source and closed-source models, and a simple pay-per-token pricing model. This allows for faster innovation with low operational complexity.
One of the key features of the Gradient AI Platform is its support for Knowledge Bases. This feature is utilized by Cloudways engineers to fetch relevant knowledge base articles alongside AI-generated insights, providing users with contextual and actionable guidance. The implementation of this feature was straightforward, with engineers ingesting knowledge base articles into the platform and exposing them through a dedicated API endpoint for seamless integration with the Copilot AI Agent.
For businesses looking to build AI agents to support critical workflows, the Gradient AI Platform offers a production-grade inference layer that powers Cloudways Copilot. The platform scales seamlessly from early experimentation to large-scale deployment, making it suitable for both early-stage startups and mature platforms. Startups can benefit from the platform’s ease of use and scalability, while growing teams can take advantage of unified model access and high availability.
In conclusion, utilizing AI agents effectively requires a strategic approach, focusing on the right tasks for AI to handle, accepting non-determinism, avoiding overestimation of AI capabilities, and being pragmatic in decision-making. The DigitalOcean Gradient AI Platform provides a reliable and flexible solution for running AI agents at scale, offering features that simplify operations and drive innovation in AI applications. Cybersecurity experts have identified a new malware strain that is targeting Android devices. The malware, named “Joker,” is capable of stealing sensitive information such as SMS messages, contact lists, and device information. Once installed on a device, Joker can also subscribe the victim to premium services without their knowledge or consent.
The malware is being distributed through malicious apps on the Google Play Store. These apps appear to be legitimate and have been downloaded by thousands of users. Once installed, Joker hides its presence on the device and begins to carry out its malicious activities in the background.
Security researchers have warned Android users to be cautious when downloading apps from third-party sources and to carefully review app permissions before installation. They also recommend using reputable antivirus software to detect and remove malicious apps from devices.
The discovery of Joker highlights the ongoing threat of malware targeting Android devices. With the increasing reliance on smartphones for everyday tasks, it is more important than ever for users to stay vigilant and protect their devices from potential threats.
Experts believe that the creators of Joker are continuously evolving the malware to evade detection and improve its capabilities. This underscores the need for constant vigilance and regular updates to security software to stay ahead of cyber threats.
In response to the discovery of Joker, Google has removed the malicious apps from the Play Store and is working to improve its security measures to prevent similar incidents in the future. However, users are advised to remain cautious and take steps to protect their devices from malware attacks.
Overall, the emergence of Joker serves as a reminder of the importance of cybersecurity in today’s digital age. By staying informed and taking proactive steps to secure devices, users can minimize the risk of falling victim to malicious attacks and protect their sensitive information from falling into the wrong hands.
For more Information, Refer to this article.
