AI Crypto News Summarizer: Per-Coin Daily Digest

Why crypto news needs an AI filter

CoinDesk, Cointelegraph, The Block, Decrypt, CryptoSlate, and the rest of Sharpe's configured source universe publish hundreds of stories per day combined. Most are repetitive, with five sites covering the same announcement, or irrelevant (analyst price predictions, sponsored content, recap articles). Reading them all is a full-time job; reading none means missing the 2-3 stories per day that actually move the market.

The middle path is a deduplicated, per-coin news feed that aggregates source articles, deduplicates similar headlines, tags coin mentions, and gives you a concise feed you can scan before opening source articles.

This page covers how Sharpe's news aggregator works: the sources, the deduplication layer, the per-coin filtering, and how to use the feed without drowning in headlines.

The 100+ source aggregation

Sharpe's news aggregator pulls from a curated list of 100+ configured RSS, Atom, Google News, and publisher feeds:

• Tier 1 publications: CoinDesk, The Block, Decrypt, Bloomberg Crypto, Reuters-linked feeds, Cointelegraph
• Crypto-native: CryptoSlate, Bitcoin Magazine, The Defiant, Protos, Unchained
• Research and newsletters: Messari, a16z Crypto, Bankless, Glassnode, Deribit Insights, Multicoin, Chainalysis
• Macro, AI, and geopolitics: Federal Reserve, ECB, OpenAI, Anthropic Blog, Google DeepMind, BBC World, Al Jazeera, Foreign Affairs

The full source list is at /news/sources. We add sources once they clear our editorial-quality thresholds: verified bylines, no pump-and-dump promotion, and no AI-generated filler.

The RSS ingestion job runs every 15 minutes. Each run normalizes source items, filters obvious sponsored content unless it is an official feed, deduplicates similar titles, tags coin mentions, and stores canonical article links in Supabase. The RSS endpoint refreshes every 10 minutes from the latest indexed rows.

How the feed reduces noise

The raw feed keeps publisher titles and links, not article text or summaries. Sharpe normalizes each indexed item into a consistent row with a headline, source, category, timestamp, and coin tags.

The feed prioritizes:

• Source quality (tier, credibility, and original reporting)
• Market relevance (price, policy, protocol, security, or liquidity impact)
• Recency (fresh stories over stale repeats)
• Topic mix (crypto, AI, markets, and geopolitics)

That structure is what a raw aggregator doesn't give you: instead of hundreds of links to click through, you start with a structured feed and decide which articles deserve a deeper read.

The per-coin filter

Each story is tagged with the coins it mentions during ingestion. The per-coin pages filter to stories relevant to that coin:

• /news/coin/bitcoin for BTC stories only
• /news/coin/ethereum for ETH stories
• /news/coin/solana for SOL stories
• /news/coin/xrp for XRP stories
• /news/coin/dogecoin for DOGE stories
• /news/coin/hyperliquid for HYPE stories

Each per-coin page ranks stories by recency after the feed resolves tickers to the canonical asset ids stored in coin_tags. The global Top news feed sits beside those coin pages as the higher-level market briefing.

The filter is automatic, with no manual curation step. Coin tagging uses explicit ticker and name matches with ambiguity filters, so common words do not become noisy token tags.

Daily scan workflow

The /news page is the daily market view. It surfaces the latest source-backed stories with category filters and search, ranked by:

• Newsworthiness (market relevance and source quality)
• Signal quality (low duplication, clear catalyst, credible source)
• Recency (within last 24 hours)
• Coin coverage (stories impacting majors weighted higher)

This is the 2-minute read. If you only have time for one news view, this is it.

How to use the feed without drowning

Three workflows:

1. Daily news scan. 2 minutes per day. Open /news, scan the latest headlines and their sources, then click into 1-2 articles if relevant to your positions. Done.

2. Per-coin alert workflow. If you hold or trade specific coins, bookmark the per-coin pages (e.g. /news/coin/bitcoin). Check once at market open and once in the evening. Most days, nothing material happens; you can skip. On days with material news, you'll see it surfaced.

3. Narrative-driven news scanning. When a narrative is rotating (e.g. AI Agents pumping), the news flow within that narrative matters. Cross-reference the narrative tracker with the news feed to find the underlying coins driving the rotation, then check news for those coins.

How this differs from CryptoPanic and LunarCrush

CryptoPanic is the closest competitor, with the same aggregation approach but without Sharpe's terminal-native coin filters. LunarCrush focuses on social sentiment alongside news. What Sharpe adds is the structured feed plus the surrounding workspace (narratives, funding rates, derivatives data) in one terminal.

Common mistakes

Treating headlines as signal. Most crypto headlines don't move markets. Always read the underlying article. A headline like "Major Hedge Fund Sells Bitcoin" might turn out to be a $50K position rather than a market-moving size.

FOMO-trading off news flow. By the time a story is in the news feed, the price has often moved. Use news for context and risk management, not for timing entries.

Ignoring sources you don't recognize. Some of the best crypto reporting comes from niche or specialist publications (Bankless, Defiant, smaller substacks). The source list surfaces these alongside major outlets, so don't filter them out.

Letting news replace fundamental analysis. News is a stream of events. A thesis is a structural view. Use news to update your thesis rather than to build it.

Where to go from here

For the daily 2-minute scan, /news is the entry. For per-coin tracking, the per-coin pages are bookmark-worthy. For deeper drilling, the main /news page supports category filters and search.

The full feed is also exposed through the public /api/news/feed and authenticated /api/v1/news/feed endpoints, which are useful for automated workflows that integrate news flow into trading or research pipelines.

There is too much crypto news to read raw. Source-quality tiers, per-coin filtering, and source deduplication cut it down to a manageable read.