Motivation

For quite some time, I've been contemplating the idea of creating my own personal search engine. I wanted a tool that could facilitate searching through my personal notes, books, articles, podcast transcripts, and anything else I wished to include. However, I was unsure of how or where to begin until I discovered the microsearch project, which reignited the momentum for the idea in my mind. I also though about it as an opportunity to delve deeper into asynchronous Python.

This project started as a clone of the microsearch project to be able to better understand how some things worked. Later, I decided to start implementing some changes like keeping all the data in a SQLite database or building a sort-of inverted index after crawling.

Features

• Fetch only what you need: winzig optimizes data retrieval by excluding previously fetched content, making sure that only new content is downloaded each time.
• Async, Async, Async: Both crawling and the subsequent data processing operate asynchronously, resulting in lightning-fast performance.
• Efficient data management: All the data is stored in a SQLite database in your home directory making it easy to retrieve and update.
• Easy to use CLI: The CLI provides simple commands for crawling and searching effortlessly, as well as clear feedback.
• Enhanced search speed: With the heavy lifting part done after fetching the content, search yields near-instant results.
• TUI: winzig also provides a basic TUI that facilitates an interactive search experience.

Installation

You'll need Python >= 3.12 to be able to run winzig.

pip

pip install winzig

pipx

pipx install winzig

Cloning this repository

Clone this repo with git clone:

git clone https://github.com/dnlzrgz/winzig winzig

Or use gh if you prefer it instead:

gh repo clone dnlzrgz/winzig

Then, create a virtualenv inside the winzig directory:

python -m venv venv

Activate the virtualenv:

source venv/bin/activate

And run:

pip install .

Instead of using pip you can also use poetry:

poetry install

And now you should be able to run:

winzig --help

Usage

To begin using Winzig, the first step is to crawl some content. The easiest method for this is to utilize the feeds file located in this repository along with the winzig crawl feeds command. These feeds will be stored in a SQLite database in your home directory, eliminating the need to provide this file again unless you're adding new feeds. If instead what you want is to crawl specific posts directly, you can use winzig crawl posts and specify a file containing the URLs you want to fetch.

Currently, there is no way to manage the feeds or posts added to the database. So if you want to remove some of them you will need to do it manually. However, it may be more efficient to delete the database and crawl again.

Crawl

The crawl command serves as a convenient and efficient method to update your database with new content. When used without any subcommands, it automatically checks for new content using the feeds stored in the database and tries to retrieves it. Basically, running:

winzig crawl

Is equivalent to:

winzig crawl feeds

Feeds

The feeds subcommand allows you to fetch and extract content from the posts of the specified feeds provided. The feeds are stored in the database so there is no need to provide a file every time.

winzig crawl feeds --file feeds.txt

winzig crawl feeds

You can also provide feed URLs directly as arguments. This feeds, if valid, will also be saved to the database.

winzig crawl feeds https://chriscoyier.net/feed/

Posts

By using the posts subcommand, you can extract content directly from the posts listed in the provided file.

winzig crawl posts --file="posts"

Or, if you prefer it, you can pass the URLs as arguments:

winzig crawl posts https://textual.textualize.io/blog/2024/02/11/file-magic-with-the-python-standard-library/

Searching

The following command starts a search for content matching the provided query and after a few seconds will return a list of relevant links.

winzig search --query="async databases with sqlalchemy"

By default the number of results is 5 but you can change this by using the -n flag.

winzig search --query="async databases with sqlalchemy" -n 10

You can add filters to your search results by using the --filter flag. Currently, the only filter supported is domain, which allows you to specify one or more domains to filter the search results.

winzig search --query "read large files" --filter domain='motherduck, textualize'

TUI

If you prefer you can use the TUI to interact with the search engine. The TUI is its early stage but it offers basic functionality and faster search experiences compared to the search command since the content is indexed once and not each time you want to search something.

winzig tui

Export

You can export your feeds and your posts to plain text or CSV format using the export command and the feeds and posts subcommands.

winzig export feeds --format csv --output feeds.csv

winzig export posts

More feeds, please

If you're looking to expand your feed collection significantly, you can get a curated list of feeds from the blogs.hn repository with just a couple of commands.

1. Download the JSON file containing the relevant information from the blogs.hn repository.

curl -sL https://raw.githubusercontent.com/surprisetalk/blogs.hn/main/blogs.json -o hn.json

2. Extract the feeds using jq. Make sure you have it installed in your system.

jq -r '.[] | select(.feed != null) | .feed' hn.json >> urls

Incorporating feeds from the resultant file will significantly increase the number of requests made. Based on my experience, fetching posts from each feed, extracting content, and performing other operations may take approximately 20 to 30 minutes, depending on your Internet connection speed. The search speed will still be pretty fast.

About the ranking function

Like the microsearch project, the ranking function used in winzig is the Okapi BM25. However, I am planning to add support for other variants of BM25, such as BM25+.

BM11 and BM15 variants

If you're using the CLI for search, you have the flexibility to adjust the k1 and b parameters. By manipulating the later to 0 or 1, you can transform the BM25 ranking function into BM15 and BM11 variants, respectively:

winzig search --query="build search engine" --b 0 # BM15
winzig search --query="build search engine" --b 1 # BM11

Roadmap

• Improve TUI.
• Add tests.
• Add multiple ranking functions.
• Add support for documents like markdown or plain text files.
• Add support for PDFs and other formats.
• Add commands to manage the SQLite database.
• Add support for advanced queries.

Contributing

If you are interested in contributing, please open an issue first. I will try to answer as soon as possible.