Cloud-based AI services are powerful, but they come with a few drawbacks. Cost, data privacy, and internet dependency are prime examples. When you’re dealing with sensitive data, need AI functionality in an offline environment, or simply want to control experimentation costs, self-hosting becomes a viable alternative. However, the process of setting up a language model on your own can be complex and time-consuming.
Ollama is an excellent open-source project that solves these problems. It allows you to run powerful language models on your personal computer with just a few commands, eliminating complex setup. It dramatically simplifies the process of downloading a specific model, running it, and integrating it with other services via an HTTP endpoint. This enables developers to focus more on how to utilize the model itself.
Code review is a core process for maintaining software quality, but it can also be a time-consuming task for team members. While AI tools like GitHub Copilot and ChatGPT have become excellent aids, security concerns about sending sensitive code to external APIs and ongoing cost issues remain.
In this article, we’ll build our own code review CLI (Command-Line Interface) tool that operates completely independently in a local environment to solve these problems. Using Ollama, a popular local LLM execution tool, and Python, we will explore how to get fast and secure feedback on your code without an external network connection.
Large Language Models (LLMs) demonstrate incredible linguistic abilities based on vast amounts of text data. However, LLMs are inherently disconnected from the external world. They cannot directly perform tasks like fetching real-time stock information, querying a database, or sending an email. The key technology to overcome this limitation is ‘Tool Use’, also known as Function Calling.
This article explains the concept and operational principles of Tool Use, focusing on Google’s Gemini 1.5 Pro model. We’ll go beyond simple API calls to delve into the tradeoffs and potential failure scenarios you might encounter in a real-world setting. This will be your first step in transforming an LLM from a simple chatbot into an agent that performs real tasks.
Countless AI coding assistants have emerged. GitHub Copilot has evolved beyond basic autocompletion to become an integral part of the development workflow, and various IDEs are rapidly incorporating their own AI features. However, most of these remain as plugins added to existing editors.
Cursor takes a slightly different approach. While based on VS Code, it markets itself as an “AI-native” IDE, designed from the ground up with AI interaction at its core. It goes beyond simple code completion to offer an experience of building software by understanding the context of the entire codebase and conversing with the user. In this article, I’ll share my experience adopting Cursor for professional work, from the setup process and impressive features to the clear limitations I encountered.
Developers perform a great deal of cognitive labor beyond just writing code. This includes analyzing legacy code, contemplating better structures for refactoring, and writing clear commit messages to explain changes. If GitHub Copilot opened the era of code auto-completion, AI tools are now becoming more deeply involved in the entire development workflow.
JetBrains AI Assistant is an AI tool integrated directly into the IDEs we use daily, such as IntelliJ, PyCharm, and WebStorm. It aims to go beyond generating simple code snippets by leveraging the rich code indexing information within the IDE to provide much more context-aware suggestions.
In this article, I will candidly discuss everything from the basic setup of JetBrains AI Assistant to its core features useful in practice, as well as failure cases and technical trade-offs I’ve personally encountered, all from the perspective of a senior developer.