🐧 OpsBuddy: A Privacy-First Local AI SysAdmin Mentor (Built with Gemma, Ollama & Sentry)

This submission for the Hacktoberfest Weekend Challenge introduces OpsBuddy, an AI-powered local mentor tailored for SysAdmin and Linux training within a command-line interface. Created for Andre, a trainee learning the fundamentals of server administration, OpsBuddy addresses the challenge of beginners grappling with deciphering intricate error logs without revealing sensitive server data (such as IPs, usernames, or paths) to cloud-based AI solutions.

By operating entirely locally on the user's machine, OpsBuddy functions as a secure, intelligent sandbox where learners can pose technical inquiries and conduct private analysis of .log files. The demo showcases OpsBuddy examining a local log file, accompanied by performance tracing data captured via Sentry. Built on a fully open-source and private stack, OpsBuddy leverages Google's Gemma 2B open-weight AI model for swift, succinct, and accurate technical guidance.

Local inference is facilitated by Ollama, running the large language model (LLM) without requiring an internet connection. The application's logic and user interface are constructed using Python and a rich library for an aesthetically pleasing CLI experience. Sentry's SDK monitors the AI's performance, incorporating start_transaction and start_child spans to accurately measure the latency of the local inference streaming.

This project exemplifies the significance of open innovation, utilizing open-weight models like Gemma 2B and open local platforms like Ollama to ensure ultimate data privacy. Users can confidently instruct the AI to evaluate sensitive SSH, database, or application logs, fearing no transmission of their data to closed, third-party APIs.

By embracing open innovation, OpsBuddy delivers a powerful, cost-free mentor that respects users' control over their information.