x2o — X Bookmark Export + AI Classification

Export X (Twitter) bookmarks → AI classify → Obsidian knowledge vault. All local, no third-party servers.

What's New (v0.5.0)

• Host AI classification: use --provider host — no API key needed, the host AI (Claude Code / OpenClaw) classifies bookmarks directly
• Standalone reclassify: pick existing output folder, reclassify with different AI
• t.co short URLs automatically resolved to real URLs
• X Notes (long-form tweets) full content extraction

Usage

Host AI Classification (no API key needed)

The host AI (Claude Code, OpenClaw, etc.) can classify bookmarks directly — no external API key required:

npx tsx ~/.claude/skills/x2o/scripts/x2o.ts \
  --cookie "<X cookie string>" \
  --provider host \
  --output ~/x2o-output \
  --limit 100

This fetches bookmarks and saves bookmarks.json. The host AI then reads the bookmarks, classifies them, and generates the Obsidian vault.

External AI Provider (with API key)

npx tsx ~/.claude/skills/x2o/scripts/x2o.ts \
  --cookie "<X cookie string>" \
  --provider openai \
  --api-key "sk-..." \
  --output ~/x2o-output \
  --limit 100

# Other providers:
# --provider claude  --api-key "sk-ant-..."
# --provider deepseek --api-key "sk-..."
# --provider gemini  --api-key "..."
# --provider ollama  --model llama3.2   (local, no api-key needed)

Re-classify existing bookmarks (skip fetching):

npx tsx ~/.claude/skills/x2o/scripts/x2o.ts \
  --input ~/x2o-output \
  --provider host \
  --output ~/x2o-output

Rebuild from Markdown vault and re-classify:

npx tsx ~/.claude/skills/x2o/scripts/x2o.ts \
  --md-dir ~/x2o-output \
  --provider host \
  --output ~/x2o-output

Parameters

Flag	Required	Description
--cookie	Yes*	X browser cookie (must contain ct0)
--input	Yes*	Path to existing bookmarks JSON or output folder (alternative to cookie)
--reclassify	No	Alias of --input
--md-dir	No	Path to a Markdown vault folder
--provider	No	AI provider: host (no API key), openai, claude, deepseek, gemini, ollama, groq, moonshot, qwen, zhipu, siliconflow, mistral, together, fireworks, xai, openrouter, cohere, deepinfra, perplexity
--api-key	No**	API key for the chosen provider
--output	No	Output directory (default: ~/x2o-output)
--limit	No	Max bookmarks to fetch (default: 800)
--model	No	Override default model for the provider
--base-url	No	Override API base URL
--fetch-only	No	Only fetch bookmarks, skip classification

* One of --cookie or --input is required ** Not required for host (uses host AI) or ollama (local)

Workflow for Claude

When user wants to export bookmarks:

1. Ask for their X cookie (guide them: F12 → Network → find any request → copy Cookie header)
2. Default to --provider host — no API key needed, you classify the bookmarks yourself
3. If user provides an API key, use the corresponding external provider instead
4. Run the script with appropriate parameters
5. For host mode: after the script saves bookmarks.json, read the file, classify each bookmark into categories (AI, Design, Crypto, Business, Career, Finance, Life, Tech, Media, News, etc.), and generate Obsidian vault markdown files
6. Report results: how many bookmarks fetched, categories found, files generated
7. Optionally open the output in Obsidian: open "obsidian://open?path=<output-dir>"

Supported AI Providers (19+)

Host AI (no API key), OpenAI, Claude, DeepSeek, Gemini, Ollama, OpenRouter, Moonshot, Qwen, Zhipu GLM, SiliconFlow, Groq, Mistral, Together AI, Fireworks, Grok (xAI), Cohere, DeepInfra, Perplexity

Source

GitHub: https://github.com/kiki123124/x2o