your home should be yours.
save everything. sort nothing. own it all.
save the link, the photo, the tweet, the note-to-self — then find it the way you think. no folders, no tags, no digging.
abode is open source under AGPL-3.0. Use the hosted version if you want it managed, eject anytime to run it yourself and own your data, free, forever.
Self-hosting is currently a work-in-progress best-effort: the code's all here and the docs are evolving, but abode's a small project, so please bear with me. The hosted version is the supported option. Issues and PRs are welcome — replies may just be slow at times.
Open-source, self-hostable app for saving images, links, tweets, videos, and articles. Visual, minimal, serendipitous; single-player first.
Your mind shouldn't be trapped on someone else's server, and you shouldn't need to commit to a lifelong subscription to access it.
Next.js 16 (React 19) · Tailwind CSS 4 · shadcn/ui · Zustand · TanStack Query · Prisma · PostgreSQL (pgvector + tsvector) · Bun · deploys on Vercel.
The only thing you must provision to self-host is a database and Supabase. Everything else is an enhancement that lights up when you add its key and degrades cleanly when you don't.
| Service | Tier | Unlocks | Without it |
|---|---|---|---|
| PostgreSQL + Supabase (auth, storage) | 🔒 Required | the app itself | won't run |
| Trigger.dev | ⭐ Recommended core | runs the enrichment pipeline and builds data exports | capture + full-text search work, but no auto-enrichment, data export or PDF import |
| OpenAI | ⭐ Recommended core | titles, descriptions, tags, OCR, semantic search | items stay bare; full-text search only |
| Replicate (CLIP) | 🧩 Optional | image embeddings (powers similar images) | skipped |
| Google Cloud Vision | 🧩 Optional | dominant colours; cheaper full-page OCR for scans and scanned PDF pages | colours skipped; document OCR uses OpenAI if configured |
| TypeSafe (Jev) | 🧩 Optional | calibrated article-vs-webpage kind refinement | structural heuristic decides the kind |
| Mapbox | 🧩 Optional | location + static map thumbnails | skipped |
| Resend | 🧩 Optional | invite / waitlist / admin emails | email features off |
| PostHog | 🧩 Optional | product analytics + session replay (saved content masked) | no telemetry (the default) |
Self-hosted instances send no telemetry unless you set your own PostHog key.
PostHog analytics and session replays are linked to the signed-in account and record how the app is used, not what people save. Everything below is masked in the browser before it's sent (app/src/lib/analytics/):
- Replays: all on-screen text, input values, images and media, and link/image URLs.
- Click events: a button's text and labels are kept only when they're UI copy written in this codebase (e.g. "Save", "Delete"); link targets are dropped.
- URLs: search queries, links shared to
/save, and the names of your own rooms (so private rooms never appear). Visits to other people's public rooms are recorded as-is.
Set NEXT_PUBLIC_SITE_URL to your instance's public URL (e.g. https://abode.example.com) so emails, share links and embeds point at your instance rather than abode.fyi. It's inlined at build time, so set it before building. Self-hosted instances are kept out of search engines (robots.txt disallows all).
To stop a single account (or an abuser) running up your AI bill, abode has durable daily/monthly per-user $ caps plus a system-wide daily circuit-breaker. They're enforced by default on any deployed instance (production, preview, or staging) — so a fresh deploy is capped without configuration and can't silently run uncapped. Local dev and tests run in shadow mode (every action counted and logged, nothing blocked); when shadow mode is active the server logs a one-time warning at first use.
Tune the thresholds to your own economics via PER_USER_DAILY_USD, PER_USER_MONTHLY_USD, and SYSTEM_DAILY_USD (all optional; sensible defaults apply if unset). To force a specific behaviour, set USAGE_LIMITS_ENFORCED=true (always enforce) or =false (deliberately opt a deploy out, e.g. to run a shadow window). Defaults are intentionally conservative — the right numbers depend on your pricing, so set your own before relying on them.
- Capture: Save via URL, file upload, paste, or text input. Supports images, articles, tweets, and videos.
- Tweets: Saved tweets keep what X shows around them: the quoted tweet, poll results, the post being replied to (or the thread it continues), and an X Article's cover, title and opening. Long-form posts — which X only shares the first ~280 characters of — are marked as cut off, with a link to the rest on X.
- Document scanning: Scan multi-page documents with your phone's camera, in the browser (no app): live edge detection, auto-capture once the page is held steady, perspective correction, and a B&W "scanned" look (or greyscale/colour). Every page is OCR'd (with OpenAI or Google Cloud Vision configured) so documents are searchable by their text, and each document is titled from it (issuer, type, date) with OpenAI.
- PDF upload: Upload, drop or paste a PDF (up to 25MB), or save a link to one, and it becomes a searchable document: pages are rendered for the document viewer (with MuPDF), the original is kept for download, and each page's embedded text is used directly, so only scanned pages are OCR'd (up to 30 per document). Titled from its text like a scan.
- Gallery: Dense masonry layout with hover actions, infinite scroll, and keyboard navigation.
- Search: Full-text search across titles, descriptions, OCR text, and extracted article content, blended with pgvector semantic (text-embedding) search via reciprocal rank fusion. Quote a phrase (
"like this") to only match items containing that exact text in their title, description, notes, tags, OCR or scanned pages (not article content). - Rooms: Manual collections and smart rooms (dynamic, filter-based), plus auto-generated book shelves (Want to read / Reading / Read) that appear once you have books and stay in sync as you update reading status.
- MCP server: Connect Claude, Cursor and other AI assistants to search and read your library (
/api/mcp, read-only), with a personal access token that has read access. - Access tokens: Personal access tokens with independent permissions: read your library (for MCP), save new items (for scripts or an iOS Shortcut), or both.
- Enrichment pipeline: Automatic metadata extraction, article parsing (Mozilla Readability), OCR and auto-tagging (OpenAI; full-page document OCR via Google Cloud Vision when configured), and embedding generation — all via async Trigger.dev tasks.
- Export: Download everything as a ZIP from Settings → Export: a complete
abode.json, a Markdown file per item (Obsidian-ready), a Netscapebookmarks.htmlwith a folder per room, a Goodreads-formatbooks.csv, and every upload, scan and saved image. Large libraries split into parts. Built in the background (needs Trigger.dev) and kept for 7 days. - Admin: User management, waitlist, and invite system.
Prerequisites: Bun, Docker (for local Supabase), and the Supabase CLI.
- Copy
.env.exampleto.envand fill in your keys — only the Required tier from External services is needed to boot. bin/install— install dependencies (runs insideapp/).bin/dev— start the dev server on http://localhost:3300 (also starts local Supabase via Docker when available).
More contributor detail — environment plumbing, port allocation, running Supabase manually, and the quality gate — is in CONTRIBUTING.md.
✅ Done (v0):
- Capture via website (URL, file, paste, compose) — images, articles, tweets, videos, products, books, notes
- Markdown notes with a WYSIWYG editor (headings, lists, checklists, quotes, code)
- Tweet context: quotes, polls, reply/thread context, X Article previews, long-form posts flagged as truncated
- Document scanning (in-browser camera, edge detection + auto-capture, B&W/greyscale/colour, multi-page, per-page OCR)
- PDFs as documents, uploaded or saved from a link (rendered pages, embedded text with OCR only for scanned pages, original kept for download)
- Masonry gallery, full-text + semantic search, filters
- Metadata extraction + article parsing (Mozilla Readability)
- OCR + auto-tagging (OpenAI)
- Dominant-colour extraction, palette bar + colour search
- pgvector text embeddings blended into search
- Similar images (CLIP visual embeddings, mean-centered)
- Location (auto + manual) + map thumbnails (Mapbox)
- Rooms (manual + smart collections)
- Public rooms, profiles + room embedding
- Article highlighting (with per-highlight notes) + shareable highlight links with quote cards
- "Save to your abode" on shared pages: signed-in readers save the link straight to their library; signed-out readers join with the sharer's invite (or, when none are available, the waitlist), and joining with an invite saves the shared link to their new abode
- Command palette (⌘K) + keyboard navigation
- MCP server for AI assistants + scoped personal access tokens
- Admin dashboard, waitlist, invite system
- Data export (JSON, Markdown, bookmarks, books CSV, every file)
🔜 Next:
- Browser extension
- Import an abode export (eject to another instance)
- Importers — bring your library from mymind, Raindrop, Are.na and Pinterest
- Self-hosting guide + Docker Compose
🔮 Later:
- Expo mobile app
- Self-hosted / privacy model option — run image analysis + embeddings without third-party APIs
AGPL-3.0 · Copyright © 2026 Jotmake Limited
Contributions are welcome. Note that abode is offered under the AGPL-3.0; a commercial license may be offered separately in the future.