A new GitHub project called Bash4LLM offers a minimalist approach to running large language models directly from the command line. Created by developer kamaludu, this single-file Bash script eliminates the need for Python or Node.js dependencies, relying instead on standard Unix tools like curl and jq that are already available on most systems.

The wrapper targets Groq's OpenAI-compatible API by default, supporting real-time streaming, file processing, and interactive chat sessions. Security considerations shaped the design: the script avoids system /tmp directories, eliminates eval usage, and implements restrictive file permissions. It also includes advanced provider validation and handles Termux environments gracefully by detecting Android and switching to atomic directory locking.

Beyond basic inference, Bash4LLM provides session management with JSON metadata, automatic output saving for large responses, and dynamic model whitelisting. Users can refresh model lists, set persistent defaults, and configure temperature and token limits through command-line flags. The modular structure separates concerns into distinct sections like PRECORE_BOOT and CORE_PROVIDER for maintainability.

Installation requires just three common packages: bash, curl, and jq. After cloning the repository and exporting your API key, you can pipe prompts directly or launch interactive sessions. The project includes optional extras for additional providers like Gemini and Mistral, though it's designed primarily for single-user environments where trust boundaries are clear.