LIVE

Psychic: A Terminal File Browser Powered by Lightweight Machine Learning

🧠 AI Analysis 2 min read 👁 0

An independent developer has released Psychic, a fast terminal file browser that uses a local LightGBM machine learning model to predict and rank files based on user habits.

Psychic: A Terminal File Browser Powered by Lightweight Machine Learning🧠AI Analysis

Photo: Hacker News Front Page

Key points

  • Psychic is a fast terminal file browser combining fzf and zoxide concepts with machine learning.
  • It uses a local LightGBM model trained on a SQLite database of user clicks and directory navigation.
  • The tool updates under 16ms per keypress and retrains from scratch during background intervals.

A new terminal utility called Psychic has been released on Homebrew, combining features of standard tools like fzf and zoxide with a localized machine learning model. Designed to streamline navigation through directories and files, the tool tracks user history and file access patterns to predict what a user wants to open next, updating search results in under 16 milliseconds per keypress.

Under the hood, Psychic stores user interactions in a local SQLite database and trains a LightGBM gradient-boosted decision tree model in the background. The developer chose LightGBM over heavier architectures like large language models or transformers due to strict latency budgets that preclude network calls or slow computations. The model retrains from scratch during background intervals, such as when a file is opened in an editor, ensuring the interface always adapts to recent usage trends.

For cold-start scenarios where interaction data is limited, the utility relies on a fallback scoring mechanism that evaluates fuzzy text matching, file modification times, and recency. As users accumulate more clicks, weight shifts dynamically toward the machine learning predictions. The tool is currently available for installation via Homebrew under the developer's repository.

What to watch next: Adoption rates among command-line power users and potential integration of similar local ranking features into other terminal productivity environments.

Source Hacker News Front Page — klauria.tech/psychic/ ↗

Contact the newsroom

Got a tip, a proposal or a partnership question? Write to us and we will reply shortly.