Aftermark

Your bookmarks are breadcrumbs.

Aftermark is a Chrome extension that treats your bookmarks as compressed intentions, not passive links. Every bookmark you saved carried an unfinished thought — something to read, compare, learn, decide, or build. Aftermark recovers that intention and helps you act on it. Local-first, privacy-respecting, no cloud dependency.

Status

v0.3.0 — Full tab UI with insights page, health scores, expanded heuristics, fuzzy duplicate detection, favicons, search upgrades, saved filters, and folder insights. All local analysis, no AI calls.

Features

Full Tab UI

Seven-view dark theme interface with sidebar navigation. Open via popup or browser action.

• Dashboard — stat cards (total, duplicates, dead, stale, average health score, clusters), top 10 domains, content type distribution, bookmarks per month, oldest/newest, bookmark rate
• All Bookmarks — searchable, sortable, filterable table with favicons, health dots, content type pills, and sort dropdown (date, domain, health, title). Saved filters for quick access
• Clusters — grouped by type with mini content-type distribution bars on each card. Click into detail view with export, rename, merge, bulk actions
• Sessions — reconstructed browsing sessions from bookmark timestamps. Expandable cards with favicons and health indicators
• Review — stale (>6 months), forgotten (>1 year), dead links, exact duplicates, fuzzy likely-duplicates, and folder insights with staleness/dead/dupe percentages
• Timeline — sparkline visualization per month (day-level bars with hover tooltips). Click to expand individual bookmarks
• Insights — bookmark personality analysis: collection health gauge, personality label (Developer/Researcher/Explorer/Shopper/Generalist), busiest month, longest drought, top domain, most-saved URL, hour-of-day histogram, day-of-week histogram, age distribution chart

Bookmark CRUD

• Add — create bookmarks from the All Bookmarks view (URL, title, tags). Auto-classifies via heuristics and syncs to Chrome bookmarks
• Edit — inline edit title, tags, and status (active/excluded) on any bookmark row. Syncs title back to Chrome via chrome.bookmarks.update()
• Delete — single delete with confirmation, or batch delete via checkboxes. Removes from both IndexedDB and Chrome bookmarks
• Bulk actions — select multiple bookmarks with checkboxes (or select-all), then Delete Selected, Mark Excluded, or Export Selected. Available in both All Bookmarks and cluster detail views
• Cluster management — rename clusters, merge two clusters, remove bookmarks from a cluster. Empty clusters are auto-pruned after any delete or removal operation

Analysis

• Full bookmark import — reads the entire Chrome bookmark tree into local IndexedDB (tested with 8,383 bookmarks)
• Content type classification — 17 types via heuristic matching: github-repo, article, video, docs, forum, shopping, travel, academic, social, reference, tool, news, real-estate, events, package, music, unknown
• Expanded domain mappings — NYT/BBC/CNN/Reuters (news), Zillow/Redfin/Craigslist (real-estate), Meetup/Eventbrite (events), npmjs/crates.io/PyPI (package), Spotify/SoundCloud/Bandcamp (music), Google Docs/Maps/Drive/Mail/Calendar
• Exact duplicate detection — smart URL normalization strips www., protocol, trailing slashes, hash fragments, sorts query parameters
• Fuzzy duplicate detection — second pass using Jaccard word-token similarity >80% within the same domain. Flags as likely-duplicate with separate review section
• Health score — 0-100 per bookmark based on age (-1/month, max -30), never revisited (-20), dead link (-30), duplicate (-15), unclassified (-10). Color-coded dot on every row (green >70, yellow 40-70, red <40)
• Dead link checker — HEAD with GET fallback, batched (10 concurrent, 500ms between batches), progress bar
• Deterministic clustering — 6 strategies: domain (5+ threshold), folder structure, time sessions (30-minute windows), URL path threads, shopping/decision groups (2-hour windows), GitHub project detection
• Session reconstruction — walks bookmarks chronologically, groups within 30-minute windows, labels by dominant domain and content type
• Folder insights — per-folder breakdown of total bookmarks, stale %, dead %, duplicate %. Dead folders (100% stale+dead) highlighted

Search & Filters

• Sort — date (newest/oldest), domain A-Z, health (best/worst), title A-Z
• Content type filter pills — click to filter, combinable with search text and sort
• Saved filters — save any search + type + sort combination with a name, stored in chrome.storage.local. Quick-access buttons with one-click delete

Visual

• Favicons — 16x16 Google favicon service on every bookmark row
• Health dots — color-coded health indicator on every row
• Sparkline timeline — day-level bars per month with hover tooltips
• Cluster mini-bars — content type distribution visualization on each cluster card

Export

• Cluster as markdown reading list
• Cluster as HTML bookmarks file (Netscape format)
• All bookmarks as CSV (title, url, domain, contentType, status, folderPath, dateAdded, tags)
• Shopping clusters as markdown comparison table
• Export selected bookmarks as CSV

Roadmap

• Milestone 4: AI enrichment — BYOK Anthropic API integration for per-bookmark summary, intent inference, and topic labeling
• Milestone 5: Semantic clustering — LLM-assisted grouping into projects, decisions, and learning paths
• Milestone 6: Artifact generation — turn clusters into reading lists, comparison tables, checklists, and research briefs

Architecture

Three-layer separation:

1. Capture/Index (local) — deterministic code handles structure, normalization, classification, clustering, health scoring
2. Inference (AI, selective) — LLM interprets meaning, summarizes, infers intent
3. Action (user-facing) — generates artifacts, surfaces recommendations

The LLM is never the source of truth. Local deterministic code owns structure, the LLM provides interpretation, and the user provides correction.

Tech

Chrome MV3, TypeScript, esbuild, IndexedDB via idb. No frameworks.

Privacy

All data stays local in IndexedDB. No telemetry. No backend. No account required. Future AI features use your own API key (BYOK) — keys are stored in chrome.storage.local and never leave your machine.

Build

npm install
npm run build

Then load the extension in Chrome:

1. Go to chrome://extensions
2. Enable Developer mode (top right)
3. Click Load unpacked
4. Select the repo root directory

License

MIT