Back to work Side Project

Glean

A personal design-knowledge tool that auto-collects inspiration weekly and uses AI to break it into readable Obsidian notes

Role
Product Definition、AI Workflow Design、Automation Development、Dashboard Design
Period
2026
Scope
AI、Automation、Node.js、Design Workflow
40s
4
Sources auto-collected (Mobbin as primary)
10
Toggleable AI breakdown sections
Two-way
Eagle ↔ Obsidian links
Weekly
Auto-scheduled + desktop notifications

Background & Challenge

As a designer, collecting inspiration is a fixed cost I pay every day: scroll Awwwards, browse new app flows on Mobbin, swing by a few sites, save things one by one. But the real problem sits further downstream — screenshots accumulate; understanding doesn’t. A folder full of great work that I never reopen three months later. What I wanted wasn’t a bigger library, but a way to read, search, and borrow the “why was it designed this way”.

So I merged two earlier side projects into one complete product: eagle-inspiration-bot (auto-collect, file into Eagle) as the Hand, and mobbin-design-digest (break down design with AI) as the Brain. Together, Glean automates the whole loop: every week it auto-collects new work from four sources → Claude breaks each down into structured notes → they’re filed into an Eagle library and an Obsidian knowledge base, linked both ways.

Design Decisions

Glean's dashboard home: a large headline 'Glean — daily inspiration, analyzed & archived', a subtitle describing weekly auto-collection from Mobbin, Awwwards and other sources, per-source toggles, and an EN/中 and light/dark toggle with a 'run now' button
The full dashboard: four sources toggled with one click, Mobbin via its official MCP for a deep breakdown, weekly scheduling and a live log — turning daily collection into a background process.
Glean's AI analysis settings card: a toggle between 'full AI analysis' and 'images only', a model picker (Haiku / Sonnet / Opus), separate outputs to Eagle and Obsidian, and a row of ten individually-checkable note sections (Overview, Key Flow, Flow diagram, Design decisions, Trade-offs, UX copy…)
The AI settings card: sections are individually checkable, the model is a three-tier choice, and persona/reader can be injected — leaving 'how deep, and for whom' to the designer.
Glean's AI note statistics: total note count, a ranked breakdown by category and design pattern, and below it recent notes where every entry has an 'Open in Obsidian' and an 'Open in Eagle' link
The knowledge base compounds: stats by category and design pattern, with every note one click from its Obsidian note or its original image in Eagle.

Outcome & Reflections

Glean is a small tool that runs locally: every week it pulls new inspiration in and breaks it into notes you can actually read, letting a library you curate yourself grow into searchable knowledge. Along the way I refactored the original 3,141-line single-file orchestrator into clean lib/ modules (bot.js slimmed to ~300 lines) and abstracted the analysis engine into a pluggable adapter layer — applying a designer’s instinct for “folding complexity into clear structure” to my own code.

But what this project really taught me is “what not to build”. Once the full version worked, I actually read each source’s terms — Mobbin (confirmed by their reply), Land-book, Awwwards, and Recent all prohibit automated scraping, and prohibit using their content for AI-derivative or commercial reuse. The problem wasn’t “which source” — it was the model itself: automated scraping + AI-derivative + outward distribution. So I made a call: this full version stays for personal use only — not open-sourced, not distributed, not turned into a product for others. A designer’s job isn’t only to ship features — it’s to see the cost clearly and know which line not to cross.