The search engine you've been using doesn't know what happened this month — and that's a problem when you're trying to understand anything that moves fast.
The Summary
- /last30days is an open-source AI agent skill that searches Reddit, X, YouTube, HN, Polymarket, and the web in parallel, then ranks results by actual engagement — upvotes, likes, real money bets — not editorial curation
- Works as a plugin for Claude Code, OpenClaw, and Hermes with zero config for most platforms; optional 30-second setup unlocks X, YouTube, and TikTok
- The edge: traditional search engines can't access most of these platforms, and individual AI models only have deals with one or two — this agent bridges the walled gardens if you bring your own API keys
The Signal
Here's what makes this interesting: every major AI assistant is blind to most of the internet that matters. ChatGPT has a Reddit deal but can't touch X. Gemini sees YouTube but not Reddit threads. Claude has none of them natively. Google Search aggregates web pages, not the comment threads where actual expertise lives.
/last30days solves this by being deliberately modular. It's not trying to be a better search index. It's an agent that coordinates access across platforms you already use, using your credentials and API keys. Reddit and HN work immediately. X, YouTube, and TikTok unlock when you add your tokens.
"Google aggregates editors. /last30days searches people."
The scoring mechanism is what separates signal from noise. Instead of PageRank or semantic similarity, it weights by engagement: Reddit upvotes, X likes, YouTube view counts, Polymarket odds backed by actual money. The agent isn't trying to predict what's important. It's measuring what millions of people already decided was worth their attention or their capital.
This matters because recency is a blind spot for foundation models. GPT-4's training data cuts off months ago. Even with web search, it's biased toward indexed pages, not real-time discussions. If you're researching a person, a company, or a technical problem, Google gives you their LinkedIn from 2023. /last30days gives you what they posted this week, what subreddit thread they're arguing in, what YouTube video explained their work better than their own website.
The architecture is simple:
- Parallel queries across platforms using their native APIs
- Agent-based synthesis that compares context across sources
- Ranking by engagement metrics, not editorial weight
- One consolidated brief instead of 47 browser tabs
It's designed to run as a skill inside existing agent frameworks. Claude Code users add it from the plugin marketplace. Hermes users drop it in their skills directory. The agent handles orchestration, you handle platform access. That's the trade: bring your own keys, get search that no single company can offer you.
The Implication
This is what agentic search looks like when it's not owned by a platform. Not a new index. Not a better model. Just better orchestration across the fragmented sources where real signal actually lives. If you're doing research, due diligence, or just trying to get smarter about something fast, this is how you'd want search to work: comprehensive, recent, and scored by reality instead of SEO.
Watch for this pattern to spread. The walled gardens won't fall, but agents that bridge them — using your credentials, your compute, your orchestration — become more valuable than any single search product. The companies that figure out agent-native search infrastructure win the next ten years.