A terminal that understands what you want to do.
An offline-first AI-powered terminal for Windows and Linux, built to make everyday shell work faster and more approachable.
What this work was about
TermiGPT grew out of a practical frustration: constantly switching between a terminal and an AI assistant to generate a command or understand an error. I built a desktop-first experience that puts shell-aware AI directly inside the terminal.
The problem to solve
Shell environments have different syntax and constraints. A useful assistant needs to work with the user’s actual shell, handle local resources privately, and make suggested actions reviewable rather than hiding everything behind automation.
How I approached it
- Built a desktop application with Electron, React, TypeScript, and xterm.js.
- Integrated on-device GGUF inference using local language models such as Qwen, Phi, and Gemma, without requiring a cloud AI API for normal operation after model setup.
- Designed shell-aware command generation and explanation workflows across PowerShell, CMD, Git Bash, and WSL.
- Added multi-tab terminal sessions, model selection, chat and error-explanation flows, plus agent-style command assistance.
- Built the accompanying public product website and release/distribution experience.
The value and the result
What it taught me
A full product from identified developer pain point to software architecture, interface design, model integration, distribution, and documentation.
Let's build something
that actually works.
Verification, automation, software products, or simply a difficult technical problem.