Hi All,
It has been about a year since I posted the first version of GoSuki here. For those who missed it, I built this tool because I was tired of bookmark managers that require a cloud account or a browser extension that tracks every page I visit.
Since v1.0, the project has evolved from a simple aggregator into a more robust local-first data tool.
What has changed since v1.0 The core philosophy remains the same—no extensions, no cloud, no telemetry—but the implementation is significantly more capable:
- P2P Synchronization (Gosuki+): I implemented a real-time peer-to-peer sync mechanism. You can now synchronize bookmarks across multiple devices without a central server or a third-party provider.
- Archival Integration: I’ve integrated ArchiveBox support, allowing users to automatically trigger local snapshots of their bookmarks to prevent linkrot.
- Expanded Browser Support: Beyond the basics, GoSuki now supports Zen Browser, IceCat, Pale Moon, and Basilisk.
- Database Resilience: The storage layer has been hardened with schema versioning and automatic backups before migrations to ensure data integrity.
Highlights of the v1.4.x Series
The latest updates focus on platform accessibility and developer workflow:
- Windows and BSD Support: GoSuki is no longer Linux-only. It now builds for Windows (with a native systray icon) and Open/Net/Free-BSD.
- macOS: Available via homebrew Tap / Cask
- Decoupled Browser Definitions: Anyone can participate to add support for new browser by defining default paths in a simple YAML file.
- Enhanced CLI: The suki command now supports advanced search and sort syntax.
Upcoming feature: Local Reddit Archival
I am currently building a deeper archival system for the Reddit module that will keep a full local copy of all your liked/upvoted posts, including comment tree and media. Lemmy would be an easy update afterwards.
GoSuki remains open source under the AGPLv3 license. If you value data ownership and want a tool that treats your bookmarks as your own, I invite you to check out the latest release.
NOTE: 100% handmade project except for a few comments and tests. I don’t even use code completion. I personnally refractored the code 3 times since inception and know every single detail of it.
It’s slop innit.
I can’t see an agents.md file, and there don’t seem to be commits signed by Claude or whatever. So I would guess this is a genuine human project?
Yeah, doesn’t look like it at first glance, but author has “LLM Ops | Bitcoin Tech” in github bio, which is icky enough on itself. And makes their project suspicious.
That stuff could be ignored in the commit, but yeah the lack of those files suggest it might not be AI-generated
Go check the commit history. All the code base is handwritten except for a few repetitive tests. I kept the original git history.
The commments are partly made with LLMs but the majority are my own.
The README is manually written only using LLM as a helper. The emojis are there to make the Readme less raw and I personally hate them but a lot of people now skip a project if it’s 100% text.
Took a look through the codebase and it seems like it could be but it’s hard to tell for sure. Some things that would lead me to believe it’s AI:
- Emojis in the readme (specifically in a bulleted list, starting each sentence).
- Big, highly descriptive, extremely consistent comment blocks for every main function.
- The shear amount of files and code and the consistency of which it was written (this could also be due to golang’s style enforcement). If written by a person, they’re beyond OCD levels of organized.
With that being said; who cares if it was made with AI? If it was: it seems thoughtfully guided by someone and it probably works fine. And if it wasn’t: this dev is nuts and just breathes production level code, no matter the project complexity haha.
Eh, emojis in the readme were in nearly every single readme template for years since before LLMs were available. I use them for my own projects just because the readme template I like has them and they make headers much more identifiable when referencing the readme, that is just human characteristics and also why electronics style guides (as an example) have standard symbols patterning for commonly used circuits. Our brains can instantly parse a familiar pattern as opposed to just text (net labels and block components opposed to shaped symbols in the example).
The shear amount of files and code and the consistency of which it was written (this could also be due to golang’s style enforcement). If written by a person, they’re beyond OCD levels of organized.
-
I started this project in 2017, I didn’t even publish the prototypes it went through .
-
I code with vim and have a script that generate template license headers in files.
-
I use LLMs to generate better comments for importent core components but if you browse the code you will find much more personal comments with obvious typos, I am not native English speaker.
Some repetitive unit tests and code sections are made with LLMs
- I make almost every commit manually as I work in a private git worktree and cherry-pick the features to the public one
Lastly I never paid for a single AI provider since ChatGPT appeared. I used the trial version of Copilot autocompletion for 2 months and never since then.
I am not anti-AI, all LLMs are self hosted with llama.cpp. I do however despise AI slop projects.
Happy to answer any questions :)
-