Agent for Teamwork Series · ClawFeed

Kevin He (@0xkevinhe)

5,000 Followings. 2 Hours Scrolling. Still Missing What Matters.

I follow about 5,000 people on Twitter.

After years in AI and Web3, my following list snowballed — researchers, founders, investors, developers, media — every follow was a “I can’t miss what this person says” decision at the time.

The result: every day, Twitter’s For You feed is a bottomless information waterfall. You can never scroll fast enough to keep up.

I tried everything:

• RSS readers → moved the problem from Twitter to another inbox
• Pocket / read-later → saved 500 articles, read 5
• Newsletter subscriptions → someone else filters for me, but their filter isn’t mine

The real problem isn’t “too much information” — it’s “too expensive to filter.” Every micro-decision of “is this worth reading?” burns attention.

It’s time for a change. Why am I using my brain to do what an Agent does best?

So I built ClawFeed 🦞 — an AI Agent that reads your entire feed, filters the noise, summarizes what matters, and delivers structured briefs. You stop scrolling. You start knowing.

ClawFeed Dashboard: 5,000 feeds → 20 curated highlights per day

AI Reads Everything. You Read What Matters.

That’s the core idea behind ClawFeed.

It doesn’t read for you. It tells you what’s worth reading. AI reads all 5,000 accounts’ feeds. You only see the highlights.

Structured Briefs, Every 4 Hours

ClawFeed’s core is multi-frequency recursive summarization:

• Every 4 hours: Extract key points from Twitter For You + Bookmarks into a structured brief
• Daily: Compress 4 briefs into the day’s 10-20 truly important items
• Weekly: Extract trends and key events from 7 daily reports
• Monthly: Generate a monthly overview for big-picture thinking

Each layer compresses the layer above it. A 5,000-person feed becomes ~20 curated highlights per day.

4-hour brief: AI extracts key signals, @username + original words format

Mark Anything → AI Deep Dive

See something worth digging into? One click, and AI generates a deep analysis — context, background, related discussions, all pulled together automatically.

Marks: flag interesting items, AI delivers full context analysis

5,000 Sources. Fully Managed.

Not just passively receiving information. ClawFeed lets you actively manage your sources:

• Smart Curation: Configurable filter rules — “prioritize AI Agent content”, “deprioritize meme coins”
• Follow/Unfollow Suggestions: AI recommends based on feed quality — some people you follow have never actually provided value

Real Results

ClawFeed after 10 days:

How It Was Built

ClawFeed didn’t start as what it is now:

v0 — Markdown + Telegram. An Agent generates a markdown summary, pushes it via Telegram. Rough but functional.

v0.2 — SQLite + API. The turning point. A real backend turned ClawFeed from a “script” into a “service.”

v0.3 — Web Dashboard. A dark-themed SPA. Finally looked like a real product.

v0.5 — OAuth Multi-user. Google OAuth so others could use it too.

Key Decisions

Zero framework dependencies. No Express, Koa, or Fastify — just Node.js native HTTP server. The only dependency is better-sqlite3. Fewer dependencies = less maintenance = fewer security risks. Under 50MB memory.

Summary format: @username + original words. “@karpathy says transformers aren’t the endgame” is far more useful than “industry discusses model architecture evolution.”

Published as both an OpenClaw Skill and a Zylos Component. One command to install on either platform. Two Agent ecosystems, one ClawFeed.

Read Less. Know More.

Information anxiety is really about filtering cost. Hand the filtering to an Agent, and the anxiety disappears.

Open Source — Full control over your data GitHub: kevinho/clawfeed ⭐ MIT license. Clone, install better-sqlite3, run.

Hosted — No signup needed, just open and go clawfeed.kevinhe.io

▶ See ClawFeed in 30 seconds:

30s demo: Hook → Dashboard → Deep Dive → Results → CTA

Read less. Know more.

Built by openclaw.ai 🦞 & zylos.ai 🐙

Twitter: @0xkevinhe