The Model Context Protocol (MCP) makes building intelligent integrations between large language models (LLMs) and cloud services easier. By combining MCP with Azure Functions, developers can expose serverless tools that LLMs can securely call in real time. This article describes how to set up Azure Functions as a custom MCP server, extend it with custom tools, and connect it to clients such as VS Code in Agent Mode.

Source

Microsoft has released Windows Terminal version 1.23 as a stable update, and version 1.24 as the latest preview, both bringing major improvements to the popular command-line tool. These versions introduce a completely new windowing architecture, expanded settings in the user interface, and other powerful enhancements for productivity and customization. In this article, we’ll explore what’s new in these updates and to how to install the latest stable and preview builds through Microsoft Store, winget, choco, and scoop.

Source

The Chorus app is built to send prompts to multiple AI models from different providers simultaneously. You can also query several search engines with each prompt. Both methods significantly reduce the risk of hallucinations.

Source

In my last post, I discussed Perplexity's aim to transform the familiar blinking web we've known for decades. It brings to mind the early days of Google, when many questioned the need for another search engine since AltaVista already did a good job. Although Comet appears to be just another Chromium-based web browser, the AI powering the browser under the hood has the same disruptive potential as Google's PageRank algorithm once did.

Source

Perplexity's new AI-powered Comet web browser, initially exclusive to Max subscribers at $200 monthly, is now accessible to Pro subscribers for $20. I tested it and am impressed—both by Perplexity's hypocrisy and Comet's truly innovative browsing experience. This post shares my rant against Perplexity, while my next post will praise Comet, the first browser on the battlefield in the new browser war.

Source

In my previous Ollama series post, I explained how to access self-hosted AI models in Ollama via HTTP. While this method provides convenient connectivity from different apps, it's only secure for on-premises Ollama hosts. For remote internet access, you can set up an SSH tunnel. Alternatively, you can securely connect from GitHub Copilot to your self-hosted AI models by using Microsoft's VS Code extension Remote - SSH.

Source

In my previous post in the Ollama series with OpenAI OSS models, I explained how to install Ollama on an Ubuntu EC2 instance. If you followed my guide, you now have everything set up to remotely access gpt-oss-20b on your self-hosted Ollama installation via HTTP. I will also showcase how to connect to remote AI models in VS Code using popular coding assistants like Cline and Roo Code, as well as through AI apps such as the local Ollama client, BoltAI, and Anything LLM.

Source

In my previous post, I covered the new features in Proxmox Backup Server 4.0. Released simultaneously, Proxmox Virtual Environment (VE) 9.0 also merits attention. It runs on Debian 13 Trixie with Kernel 6.14 and includes updated QEMU, LXC, ZFS, and Ceph versions. More significantly, it introduces modifications affecting storage management, networking, and high availability within your virtual setup. This article examines these updates more closely, providing commands and examples you can test on your own system.

Source

In my previous post, I explained why you might want to run OpenAI gpt-oss-20b or gpt-oss-120b on your own virtual machine and how to select an appropriate EC2 instance type. Today, I will show you how to install Ollama via script or snap on Ubuntu within an EC2 instance. Most steps can be adapted for other cloud providers, on-premises setups, or Linux distributions with minor modifications. I will also briefly discuss the benefits of running open-weight models on an EC2 instance with Ollama compared to Bedrock, SageMaker, vLLM, or Hugging Face Transformers.

Source

OpenAI recently released its first open-weight models, gpt-oss-20b and gpt‑oss-120b, which you can download and run with Ollama. If security and privacy are essential when using AI models, you might not want to use the vendor's chat UI or API. The most powerful AI models require significant computing power and memory, and many desktop devices cannot run these models. Although Ollama was not designed with remote connections in mind, it is still possible. In this series, I will explain how to install OpenAI's open-weight models using Ollama and connect to them remotely from various applications, including VS Code GitHub Copilot. In this first post, I will outline how to choose the right size for your virtual machine, using EC2 instance types in AWS as an example.

Source

Proxmox Backup Server (PBS) 4.0 is here with a new Debian 13 base, updated kernel, native S3-compatible storage, and useful improvements for managing backups. This release gives you more choices for where to store data and simplifies tasks like using removable drives or syncing to off-site locations. In this post, I will walk you through crucial new features, show the commands you can run, and provide a short upgrade checklist.

Source

The key new feature in the just-released AI Shell Preview 6 is run_command_in_terminal, among several new built-in tools that allow the AI model to run commands and review outputs directly in your PowerShell terminal. Additionally, AI Shell now supports local and remote MCP servers, transforming it into a versatile AI assistant across Windows, Mac, and Linux.

Source

The ChatGPT Gmail connector was first released in June 2025 for Plus, Pro, Team, Enterprise, and Edu users. However, because of alleged data protection rules, users in the European Economic Area, Switzerland, and the UK were excluded. OpenAI is now deploying connectors for Gmail, Google Calendar, Google Drive, and GitHub across Europe for all plans—notably after Anthropic released their MCP connectors weeks ago in Europe without such concerns, suggesting OpenAI's initial data protection claims were merely pretextual or that MCP has no priority status for the AI company. The announcement mentions Google Contacts, but this connector isn't yet accessible in my ChatGPT with a Plus plan. I tested the Gmail Connector, and following GPT-5's disastrous launch, it appears OpenAI released another flop immediately afterward.

Source

If you are working with PowerShell, you will often face the question: "Under which user is my script running?" In other words, you want to know who is logged in. Many admins use the whoami command on the Command Prompt (cmd). However, PowerShell provides native alternatives to whoami that give you the username and domain.

Source

The web is flooded with complaints from GPT-5 users, prompting OpenAI's CEO to reintroduce GPT-4o as an alternative. This marks the third setback following the departure of chief scientist Ila Sukestver, who led GPT development until version 4. One likely cause of user frustration is that, following Altman's recent boasts, everyone anticipated a major advance towards "AGI," but we are merely observing another crash into the AI Wall.

Source

In my previous post, I covered Microsoft’s new Edge Copilot Mode, introducing only one significant change. However, shifting Copilot from the sidebar to the center in Edge is more than just a design update. Microsoft’s move marks the first strategic step in the unfolding new browser war involving Google, OpenAI, and Perplexity—companies that exploit broad 'fair use' claims to harvest web content for their own greedy gain. But the upcoming battles aren’t just about who controls the web’s entry point—it could signal the beginning of the end of the web as we know it. The fall of Google’s dominance as the web’s central policymaker now seems inevitable. Once the dust settles from the new browser war, a decentralized web shaped by independently communicating AI agents might emerge. Think that’s far-fetched? Keep reading.

Source

The Dynamic Host Configuration Protocol (DHCP) server handles automatic IP address assignment within a company network. This article explains how to transfer the DHCP server role to Windows Server 2025 using PowerShell, covering the export and import of DHCP scopes, leases, and reservations to facilitate a seamless migration with minimal disruption to users and systems.

Source

OpenAI has just released its first open-weight models, gpt-oss:20 and gpt-oss:120, enabling you to run these surprisingly powerful models locally, which benefits security and privacy. I downloaded gpt-oss:20 and tested it on my laptop using Ollama and within VS Code with GitHub Copilot. To my surprise, the 20 billion parameter model gpt-oss:20 responded to my prompt.

Source

In Edge Copilot Mode, announced a few days ago, Copilot shifts from being just an AI assistant to becoming the main browser interface. Now front and center in Microsoft's browser, Copilot has evolved into a serious ChatGPT competitor. In this post, I explain how to enable Edge Copilot Mode and discuss whether it introduces any new features compared to the previous version of Copilot in Edge.

Source