What it does Siftly pulls your Twitter/X bookmarks into a local Next.js app and subjects them to a small industrial process: entity extraction, vision analysis of every image and thumbnail, semantic tagging, and AI categorization. The result is a searchable, filterable personal archive with an interactive mindmap view and natural-language search that reads tweet text, OCR’d images, and generated tags all at once.

The interesting bit The zero-config auth is a neat trick: if you already use Claude Code CLI on macOS, Siftly reads your session from the keychain and just starts working. No API key, no env vars, no billing setup. For everyone else, it falls back to standard Anthropic/OpenAI/MiniMax keys. The import flow is similarly pragmatic — a bookmarklet or console script that auto-scrolls your X bookmarks page, no browser extension required.

Key highlights

• Local-first: SQLite database, no cloud storage, AI API calls are the only external traffic
• 4-stage pipeline: free entity extraction → vision analysis (30–40 tags per image) → semantic tagging (25–35 tags) → categorization with confidence scores
• FTS5 full-text search + Claude semantic reranking for natural-language queries
• Export to CSV, JSON, or ZIP with media files and manifest
• Supports Anthropic, OpenAI, and MiniMax models; custom base URLs for local proxies

Caveats

• Claude CLI auto-auth is macOS-only; Linux/Windows users need an API key
• The README is thorough but the architecture section trails off mid-sentence, suggesting the docs may be slightly ahead of the repo or just incomplete

Verdict Worth a spin if your X bookmarks have become an unsearchable graveyard of half-remembered threads. Skip it if you don’t save tweets or already have a note-taking system you actually maintain.