Slow and Steady: local, open-source AI that helps you work without urgency or cloud lock-in

This submission is for the Hacktoberfest Open-Source AI Challenge Week 3 with the theme "Slow and Steady." The project is called "Slow and Steady," a personal AI workspace that runs on the machine it's installed on. The developers chose the name sloth-inspired to emphasize the app's low-key approach. Most "AI assistants" require users to create an account, paste private text into an external API, and pay per token.

However, this app takes a different approach. After installation, users can interact with an open-source model directly on their local machine, keeping conversations in a local SQLite file. The app is designed to assist with tasks people often postpone, such as explaining difficult topics, refining paragraphs, sketching projects, walking through code snippets, or simply providing a calm chat environment.

Users have the option to use a demo mode, which offers scripted replies, or connect to a real open-weight model installed in Ollama. The app features streaming, stop, regenerate, copy, Markdown, code highlighting, and conversation history, all functioning similarly in both demo and real model modes. The developers have built two processes running on the local loopback: a browser interface using React, TypeScript, Vite, Tailwind CSS, and Lucide, and a FastAPI backend on 127.0.0.1:8747 connected to Ollama on 127.0.0.1:11434.

All data, including chats and settings, are stored in a local SQLite database. The UI never communicates with the model directly. Instead, an abstract AIProvider sits behind the API, implementing two specific providers: DemoProvider, which provides scripted replies, and OllamaProvider, which communicates with Ollama using local HTTP only, without any cloud fallback or automatic model downloads.

The project follows a minimalist security approach, binding the API to localhost, limiting CORS to the local Vite origin, and validating various parameters, such as temperature, token limits, message length, and model names. The developers emphasize that the app is designed to be energy-efficient, allowing users to explore its features even at 11 PM without waiting for a lengthy model download.

When users are ready, they can connect a real Ollama model and continue chatting. The project adheres to the spirit of open innovation, ensuring that users' words remain private and that the model runs locally on their machine without any cloud lock-in.