Read Later
Distillation at your own pace. Capture articles, videos, and notes into clean markdown, read anywhere offline.
Read Later list view, light/dark theme.
The focused reading list. Convert noisy web pages and notes into clean, offline-ready markdown.
Read Later is a lightweight, self-hosted reading platform designed to eliminate clutter and save articles, YouTube videos, and web pages for offline reading. Unlike commercial bookmarking services that lock content behind proprietary reader formats, Read Later extracts raw HTML into clean, standardized Markdown saved directly in PostgreSQL.
Content capture spans multiple channels: a zero-dependency browser bookmarklet for desktop and iOS Safari, native PWA Web Share Target for Android, and an iOS Shortcut—all routing through a unified Supabase edge function. Saved links trigger a background worker that fetches and parses content, while free-form text shares instantly bypass worker queues as direct notes (type='note', status='ready').
The system includes weighted Postgres full-text search, create-on-type relational tagging, unguessable public link sharing (/s/:id), and atomic URL deduplication.
Translation of the content - by URL (articles added - Google Translate web-site) or inline (for snippets - DeepL API).
Timeline
Q2 · 2026
Scope
Product Architecture
Vue 3 PWA Reader
Parser Worker
Search & Tagging Engine
Scale
Personal Archive
Multi-device Sync
Decoupled components for zero friction and total privacy.
Engineered for simplicity and durability.
A single-line JavaScript bookmarklet works seamlessly across desktop browsers and iOS Safari without background battery drain or browser store approvals. Combined with a Web App Manifest share_target for Android and iOS Shortcut integration, all capture channels share a single static bearer key on the Supabase capture edge function.
Markdown content is indexed directly in PostgreSQL using a generated tsvector column (search_vector) with weighted titles (setweight 'A') and body content (setweight 'B'), indexed with GIN (bookmarks_search_idx). Client queries debounce 300ms, and gracefully degrade to offline title/excerpt substring matching when disconnected.
Replaced flat array storage with a relational tags and bookmark_tags schema. Creating indexes with tag_id leading ensures sub-millisecond tag-based filtering (search_vector @@ query AND tag_id in (...)), while create-on-type upserts avoid client-side race conditions.
Public sharing uses unguessable UUID URLs (/s/:id) exempted from the PWA’s auth route guard. Instead of a blanket is_public = true RLS policy—which allows anyone with the publishable key to enumerate all public bookmarks—reads are restricted to a security definer RPC function (get_public_bookmark) requiring the exact bookmark ID.
A generated url_normalized column (lowercased host, stripped trailing slashes) with a unique index (bookmarks_url_normalized_uniq) enforces duplicate prevention at the database level. Conflicts return a duplicate status, allowing users to cancel or re-trigger a background parse (POST /capture/:id/refresh).
Capturing non-URL text automatically creates a direct note entry (type='note', status='ready'). This bypasses the VPS worker queue entirely, giving instant zero-latency availability for personal snippets and thoughts.
Distillation at your fingertips.
Read Later delivers a fast, private, and fully customizable reading archive. The updated architecture seamlessly combines multi-channel capture, instant full-text search, relational tagging, unguessable public link sharing, and offline PWA capabilities—providing a clutter-free environment for deep reading and knowledge retention.
Impact
Zero reading distractions
Instant full-text search
Relational tags & notes
Unguessable public links
Offline PWA support
Next
Inline note editor
URL query tracking stripper
Prerendered Open Graph cards
Reading article.
Reading progress.