X (formerly Twitter) bookmarks get harder to find the more you save. Siftly imports that backlog, analyzes it with AI, attaches tags and categories, and turns it into a knowledge base you can search in plain language. It is developed by Viperr. The Next.js web app is self-hosted — you run it on your own machine, all data lives in a local SQLite database, and there is no browser extension to install and no upload to a vendor’s servers. The first release shipped in March 2026, followed by OpenAI support, Obsidian export, and an AI category assistant. Licensed under MIT.
Key Features
- A four-stage AI pipeline: Imported bookmarks pass through 1. entity extraction (hashtags, URLs, mentions, and known tool names pulled straight from the tweet JSON, with zero API calls), 2. vision analysis (reading text and subjects out of images, GIFs, and video thumbnails), 3. semantic tagging (generating searchable tags from the post text combined with the image findings), and 4. categorization (assigning one to three categories with confidence scores)
- Natural-language AI search: Vague phrasing like “the post with that chart in it” still lands on the right bookmark, because the search leans on the generated tags and categories rather than exact keyword matches
- Interactive mind map: A force-directed graph lays bookmarks out by category, giving you a bird’s-eye view of what you have actually been saving. It is a different way to look than scrolling a list
- Fully local storage: Your bookmarks live in a SQLite database on your machine, with FTS5 full-text search. There is no telemetry, and you never create an account with Siftly
- Multiple AI providers: Choose from Anthropic (Claude Haiku 4.5 / Sonnet 4.6 / Opus 4.6), OpenAI (the GPT-4.1 family, o4-mini, o3), or MiniMax (M2.7 and others). If you already use the Claude Code CLI, Siftly detects that session and starts analyzing without any API key setup
- Export options: CSV (for spreadsheets), JSON (the full structured dataset), and ZIP (bookmarks by category with media files and a manifest), plus Markdown export shaped for Obsidian
Pricing
| Plan | Cost | What it covers |
|---|---|---|
| The app itself (OSS) | $0 | MIT licensed. No feature limits, no subscription |
| AI analysis | Usage-based (paid to your AI provider) | API charges for the model you pick. No extra API key needed when using the Claude Code CLI |
Pricing is accurate as of August 2026. Siftly itself is free, but the vision-analysis stage onward incurs usage fees from whichever AI provider you choose. Check the official repository for the latest details, and each provider’s pricing page for model rates.
Pros & Cons
✅ Pros
- Your bookmarks stay on your own machine, so a shutdown or policy change at some vendor cannot take the data away
- No subscription — you only pay the real cost of AI analysis, and nothing at all when you are not analyzing
- No browser extension required, which makes it easier to adopt if you would rather not grant an extension access to your browsing
- Obsidian export lets you feed the bookmarks into a note-taking setup you already use
- If you already run the Claude Code CLI, you can skip the API-key issuing steps entirely
⚠️ Cons
- Being self-hosted, it requires Node.js 18 or newer and a setup pass of your own. There is no “open a URL and go” convenience of a hosted service
- “Fully local” describes where data is stored, not the whole pipeline: the vision, tagging, and categorization stages send content to the AI provider you selected. It is not an offline tool by default (a custom endpoint setting is available for local models)
- It handles X bookmarks only. It is not a general read-it-later tool for other platforms
- Version numbering is inconsistent (v1.0.1 came before v0.2.0 and v0.3.0), so maturity is hard to judge from the version alone
- It is a young solo project, with no guarantee of long-term maintenance
Comparison with Similar Services
| Criteria | Siftly | Raindrop.io | Readwise Reader | Obsidian + Web Clipper |
|---|---|---|---|---|
| Delivery | Self-hosted (OSS) | Cloud service | Cloud service | Local app + extension |
| Main target | X bookmarks | Web pages generally | Articles, PDFs, tweets | Web pages generally |
| AI classification | Automatic four-stage pipeline | AI suggestions (paid plans) | AI summaries and highlights | Not built in |
| Where data lives | Local SQLite | Vendor servers | Vendor servers | Local Markdown |
| Cost | Free plus AI usage | Free tier / paid plans | Paid | Free / paid (sync) |
Who Is It For
- Anyone sitting on hundreds or thousands of X bookmarks they can no longer dig through
- People who want their personal archive on their own machine rather than entrusted to an outside service
- Obsidian (or similar) users who want their X saves consolidated into the same notes
- Developers already using the Claude Code CLI or AI APIs who are comfortable setting up a local tool
- Anyone who would rather pay per use than add another monthly subscription
Summary
Siftly takes X bookmarks — an asset most people accumulate and never revisit — and makes them retrievable through AI tagging and natural-language search. Its strengths are keeping data local, requiring no subscription, and exporting to Obsidian; the trade-off is that preparing a Node.js environment and running the initial setup is on you. As long as you understand that the analysis stages hand content to your chosen provider, the sensible first step is to import the bookmarks you already have and see how much of them you can suddenly find.