GitHub Repository

A smart picker for Snacks.nvim that trains a neural network with your file picking preferences.

62 starsLua

Neural-open.nvim – Neural network powered Neovim file picker

Name: Neural-open.nvim – Neural network powered Neovim file picker
Availability: InStock
Author: dtormoen

by dtormoen·Feb 26, 2026·2 points·0 comments

Visit Project View on HN

AI Analysis

●●SolidNiche GemCozy

MLP learns your nav patterns—clever, but Telescope and fzf already solve file finding.

Strengths

•Ships with pre-trained weights from 16k samples, so ranking is useful immediately without cold-start phase.
•Trains online on your selections using pairwise hinge loss; persists weights across sessions for continuous adaptation.
•11 scoring features (fuzzy, frecency, transitions, directory proximity) meaningfully richer than basic fuzzy matching.

Weaknesses

•Audience is tiny: requires Snacks.nvim, Neovim 0.11+, and active development workflow using file picker multiple times daily.
•No benchmarks showing whether neural ranking actually beats smart-open.nvim or other learned-ranking pickers in practice.

Post Description

neural-open.nvim is a Neovim plugin that integrates with Snacks.nvim that trains a neural network to rank files based on your navigation patterns using features like fuzzy-finder match scores, file access frecency, and directory structure. It trains a small multi-layer perceptron and updates the network for a few iterations each time you choose a file to learn your preferences.

I made this because I always felt other file search solutions never fully match my flow and could learn more from how I use them to be better. It was also a fun little side project to help me remember some of the details of the math in deep learning that had gotten a little stale over time.