I've been experimenting with Google Cloud Platform (GCP) to assemble a personal toolset. Lately, the volume of information I've been juggling has left me feeling overwhelmed. To tame that, I've turned toward personal knowledge management platforms like Obsidian.
However, my preferences are a bit tricky: I want open-source options, I want to own my data, and I think it's a good idea to have it all structured in a way that's digestible by large language models (LLMs). That combination led me to Logseq.
The Logseq Dilemma
Logseq checks a lot of boxes in terms of features, but one issue stands out: it's local-first. I want to maintain a synchronized graph across my phone, personal computer, and, on occassion, devices I don't own (no installations or use of file system for my own graph). Unfortunately, Logseq’s built-in sync is weak, and there's hosted front end uses the local filesystem as the data base.
Database Storage
Logseq's database is straightforward: a collection of cross-referenced, human-readable Markdown files. Perfect. That makes an S3-compatible bucket an ideal backend. Since I'm leaning into GCP, I decided to use a Google Cloud Storage bucket.
Syncing Across Devices
Laptop
For the laptop, mounting to an S3-compatible bucket is simple with rclone. It handles the file synchronization reliably.
Smartphone
On Android, I used FolderSync to connect and authenticate with the GCP bucket. Syncing runs on a schedule—every hour feels like a good compromise between battery consumption and freshness. Ideally, it would be event-driven file watching, but for now this serves as a minimum viable product (MVP).
Hosted Access on the Go
The final challenge is being able to access Logseq from any arbitrary, internet-connected machine. That requires a hosted instance of Logseq, premounted to the bucket database.
Since this project only has a single user, running a 24/7 VM isn't sensible. Instead, the compromise is a lightweight, Dockerized remote desktop. For this, I used accetto's noVNC Docker image deployed with Cloud Run, mounted directly to the storage bucket.
This setup allows me to run Logseq in the browser, without downloading software or exposing my entire, personal knowledge graph.
Logseq implementation Architecture
Comments
Post a Comment