How I Built an Offline AI Assistant in Python - No OpenAI, No LangChain, No Dependencies
SMRTR summary
A solo developer, building software between retail shifts, got tired of AI assistants that die without internet, cost money per message, and send your data to the cloud.
So Marcin Firmuga built something different.
His system, hck_GPT, runs entirely offline, handles both English and Polish without any language detection library, and routes every question through a five-step decision chain that blends keyword scoring with machine learning confidence scores.
The cleverness is in the details. The assistant adjusts its "creativity level" depending on what you're asking. A question about CPU specs gets a precise, low-temperature response. "How are you?" gets warmth and personality. It also remembers what it said three messages ago, and if the local AI model crashes or isn't installed, it falls back to a rule-based engine in milliseconds.
No cloud. No API fees. No spinner while a server thinks.
Firmuga built it for PC Workman, an open-source system monitor, and published everything on GitHub, 800 hours of work, built publicly, between retail shifts.
SMRTR provides this summary for quick context. The original article belongs to Dev.to.
Read the original article