♦︎ ETH24

Daily Ethereum intelligence. The top tweets from the last 24 hours, ranked by importance with one-line commentary.

Fork it for any topic.

How It Works

1. Crawl - Dual Grok x_search (trending + ecosystem) finds tweets by context, dual X API v2 search (relevancy + recency) finds tweets by engagement, RSS feeds for background context
2. Rank - AI selects up to 10 by importance, writes one-line commentary
3. Output - CLI (stdout) or tweet mode (formatted post preview)

Grok x_search (trending) ──\
                             ├──> contextual discovery
Grok x_search (ecosystem) ──/            \
                                    X API lookup ──> engagement metrics
                                          /                       |
X API search (relevancy) ──\             /                   merge + dedup
                            ├──> keyword tweets                   |
X API search (recency)  ──/                             RSS ──> AI ranking
                                                                  |
                                                            ♦︎ output

Why This Architecture

Neither data source alone is good enough.

Grok x_search understands context. It finds tweets that matter even if engagement is low - a new EIP with 50 likes can be more important than a meme with 5,000. But Grok doesn't return reliable engagement metrics.

X API v2 has real metrics (likes, retweets, replies) but finds tweets by keyword matching, which surfaces a lot of noise. Popular doesn't mean important.

The hybrid approach: Grok discovers what's important. X API verifies how much traction it has. The merge gives you both signal and validation. Either source can run independently if the other fails - Grok-only or X API-only both produce usable results, just less complete.

Why the ranking is a separate AI step: Engagement scores alone produce bad digests. A spam airdrop tweet with botted likes would outrank a meaningful protocol update. The AI ranking step filters junk, deduplicates overlapping stories, and writes commentary that adds context the original tweet doesn't provide.

No hardcoded account list - anyone can surface a top tweet. Candidates are scored by (likes + retweets) / 2 with a configurable engagement floor that auto-lowers on quiet days.

Cost Per Run

Step	API	Cost
Grok discovery (2 queries)	xAI Responses API (grok-4-1-fast + x_search)	~$0.30
X API search (2 passes)	X API v2 recent search (up to 600 tweets)	~$3.00
X API lookup	X API v2 tweet lookup (10-50 enrichments)	~$0.10
AI ranking	Anthropic Claude or xAI Grok (fallback)	~$0.10
Total per run		~$3.50
Monthly (daily)		~$105

When using the Claude Code skill flow (/eth24), the ranking step is free - Claude handles it directly. Drops the per-run cost to ~$3.40.

X API is pay-per-use at ~$0.005/tweet read. The xAI minimum credit buy is $25, which covers ~130 daily runs of the Grok steps alone.

Install

Option A: ClawHub

npx clawhub@latest install eth24

Option B: Claude Code Skill

Clone this repo and run /eth24 from Claude Code. Claude handles the ranking directly - no separate AI API key needed for that step.

Option C: Standalone

pip install -r requirements.txt
python3 main.py            # CLI mode (default)
python3 main.py --mode tweet  # Tweet-formatted preview

Setup

API Keys

XAI_API_KEY=...           # xAI API key (console.x.ai) - Grok discovery + ranking fallback
X_BEARER_TOKEN=...        # X API v2 bearer token (developer.x.com) - keyword search + metrics
ANTHROPIC_API_KEY=...     # Anthropic API key (optional, primary ranking provider)

At minimum, set XAI_API_KEY and X_BEARER_TOKEN. Add ANTHROPIC_API_KEY for Claude-based ranking (falls back to Grok if unavailable).

Configure Your Topic

Edit config.json:

{
  "topic": "Ethereum",
  "brand": {
    "name": "ETH24",
    "account": "@yourhandle",
    "repo_url": "github.com/you/your-fork"
  },
  "crawl": {
    "x_search_terms": ["Ethereum", "ETH", "EIP", "Pectra", "L2", "rollups"],
    "rss_feeds": {
      "The Block": "https://www.theblock.co/rss.xml",
      "Bankless": "https://feeds.banklesshq.com/rss"
    },
    "lookback_hours": 24,
    "max_tweets": 10,
    "engagement_floor": 200,
    "max_candidates": 50
  },
  "rank": {
    "ai_provider_order": ["anthropic", "xai"],
    "voice": "Concise, informed, direct. Short sentences. Plain language. No emojis. No emdashes."
  }
}

Fork It

ETH24 is built for Ethereum but the pipeline is topic-agnostic. To run it for something else:

1. Fork this repo
2. Edit config.json: change topic, brand, x_search_terms, rss_feeds
3. Set your API keys
4. Run it

Examples: SOL24, AI24, DEFI24, BTC24 - same architecture, different search terms.

Output

ETH24 - 2/18/26

Vitalik on surveillance, EIP-7623 efficiency, privacy momentum

♦︎ Vitalik underscores the real-world harms of surveillance,
  highlighting privacy's urgency in Ethereum.
  https://x.com/VitalikButerin/status/2024106047460475063

♦︎ EIP-7623 increases calldata gas costs but keeps overall
  expenses low due to other optimizations.
  https://x.com/nero_eth/status/2023681491499163911

All output goes to output/YYYY-MM-DD/:

• crawled.json - raw tweet data with engagement metrics
• ranked.json - selected stories with commentary
• cli.txt - formatted plain text (CLI mode)
• thread.txt - formatted tweet (tweet mode)

Credits

Built by Pat McGowan at Number Group for the Ethereum Community Foundation.

MIT License