What is Parrot?
Parrot is an AI assistant that runs locally on your desktop. It leverages local LLMs powered by Ollama to provide a completely private experience. Parrot's AI assistant is deeply integrated into your desktop environment and is designed to make use of local resources efficiently to deliver tailored responses. We achieved this by feeding context from your filesystem indexes, applications, system information, and more. Furthermore, Parrot is designed to automatically fetch relevant context based on your queries, ensuring that the responses are accurate and personalized.
Features
- Local LLMs powered by Ollama's API
- Lightweight and beautiful desktop application built with Tauri
- Deep integration with your desktop environment
- Automatic context fetching from your filesystem, applications, and system information
- Settings: auto-execution and user-approved execution of commands
Technologies
How it works
Parrot is built using Tauri, a Rust framework for building desktop applications with web technologies. The frontend is developed using Next.js, allowing us to rapidly design the frontend with React components. The backend is powered by Rust, which allows us to efficiently interact with the operating system and fetch context from various sources.
When Parrot receives a query, the query is sent to the Ollama API with a predefined amount of context including general system information and installed programs. The LLM is given a set of commands it can use to either execute actions on the users desktop or fetch more context. If more context is needed, Parrot will automatically fetch the relevant information and send it back to the LLM. Once the LLM has enough context, it will generate a set of commands to execute on the user's computer. Depending on the user's settings, these commands can either be executed automatically or require user approval.
My contribution
I was primarily responsible for the backend of Parrot, which involved setting up the Tauri application, integrating with the Ollama API, and implementing the logic for fetching context and executing commands. I also contributed to the frontend by designing and implementing several React components and improving the overall user experience.
Inspiration
The inspiration for Parrot came from my personal experience of using AI assistants like ChatGPT and noticing the limitations of cloud-based LLMs in terms of privacy and context-awareness. I wanted to create a solution that would allow users to leverage the power of LLMs on their own local machines while ensuring their data remains private.
Parrot was a hackathon project and is no longer being actively maintained.