Tutorials
In the evolving landscape of software development, the integration of artificial intelligence into daily workflows has become increasingly prevalent. A notable recent development is the emergence of Crush, a Go-based command-line interface (CLI) application that brings AI capabilities directly into your terminal. This tool, previously known as OpenCode AI, underwent a rebranding and ownership change on July 29, 2025, signaling a fresh chapter in its journey to provide developers with sophisticated AI assistance.
Crush stands out by offering a robust Text User Interface (TUI), enabling seamless interaction with a diverse array of AI models. It's designed to streamline various coding tasks, facilitate debugging processes, and offer general support, effectively positioning itself as a developer’s 'coding bestie.'
The underlying technology for Crush’s TUI is the acclaimed BubbleTea project, known for its elegant design in crafting interactive terminal applications. Crush itself is built to be multimodel and session-aware, allowing users the flexibility to select from numerous large language models (LLMs) or even integrate their own. A particularly useful feature is the ability to switch between LLMs during an active session without losing conversational context, maintaining a fluid development experience. Furthermore, it supports the integration of custom Language Server Protocol (LSP) configurations, akin to how one would set them up in a standard code editor.
Overview of Crush's Capabilities and Practical Application
Getting started with Crush is straightforward, particularly for macOS users who can leverage Homebrew for installation. The application guides users through the initial setup, including key management for chosen LLM providers. For instance, developers can configure their preferred LLM, such as those from Anthropic, by setting environment variables, ensuring secure and personalized access to AI services.
Crush supports a wide range of LLMs, encompassing both popular choices like Opus 4 and GPT-5, as well as lesser-known but powerful alternatives such as Groq. Upon first launch, Crush prompts users to initialize its configuration, creating a context file that helps tailor the AI’s responses to the user’s project. This interactive setup process, though employing an informal tone, simplifies the onboarding experience for developers.
One of Crush’s distinctive features is its meticulous approach to permissions. When executing operating system commands from within Crush, such as navigating directories, the application proactively seeks user confirmation. This behavior is governed by a configurable crush.json file, where developers can define an explicit 'allow' list for tools like 'view', 'ls', 'grep', and 'edit'. This granular control ensures that AI actions are transparent and sanctioned, though a --yolo flag exists for those who prefer to bypass these checks, indicative of the project's somewhat playful yet powerful design philosophy.
Real-World Testing and Insights
To assess its practical utility, Crush was tested on a Unity development directory. The application efficiently processed the project, analyzing configurations, tests, and project files. It also intelligently looked for existing instructions from other AI coding assistants like Cursor or Copilot to integrate them, suggesting a future where it might support a broader spectrum of provider instruction files. A thoughtful addition is the quit command in the command line, enhancing user convenience, although some terminal emulator specific issues were observed, such as limitations with mouse interaction and text copying within certain terminal environments.
During one test, after encountering a token limit with the Anthropic API, the configuration was refreshed. Upon reprocessing, Crush successfully generated a concise CRUSH.md file, which was then automatically added to the .gitignore, demonstrating its ability to manage project documentation effectively.
Further testing involved tasking Crush with generating a new class within a collection of strategies for a game project. The AI adeptly interpreted the context, such as identifying 'farmer' as both a character name and a type, and intelligently refined variable names, showcasing its contextual understanding. Critically, before making any modifications, Crush requested explicit permission, illustrating its commitment to controlled and transparent operations. While it made a commendable attempt to add instances of the new narrative, it recognized when it lacked sufficient information, allowing the user to halt the process at an appropriate juncture.
Although Crush currently presents some user interface aspects that could benefit from further refinement, its innovative features and consistent TUI approach position it as a formidable tool in the AI-assisted development space. While it might not yet surpass the integrated capabilities of some advanced terminal environments, its open-source nature promises continuous enhancements and a growing set of robust functionalities, making it a compelling option for developers seeking AI support directly within their terminal workflows.
From a developer's perspective, Crush introduces a fascinating blend of AI power and terminal simplicity. Its commitment to transparency through explicit permission requests for file operations is a crucial step towards building trust in AI-driven development tools. As AI continues to embed itself deeper into our coding practices, tools like Crush underscore the importance of maintaining user control and contextual awareness. The ability to switch LLMs mid-session while preserving context is a game-changer, allowing for flexibility and experimentation without interrupting the flow of thought. While it has areas for growth, particularly in UI polish and seamless integration with all terminal emulators, Crush's foundational design points to a future where AI is not just a helper, but an interactive and intuitive partner in every line of code we write.