khoj — Your AI second brain. Self-hostable. Get answers from the web or your docs. Build custom…

1. Overview and Visual Introduction

• The project is titled last30days-skill, an AI agent-led search engine designed to surface what real people engage with across multiple platforms, not what editors claim or SEO signals push.
• At the top, two visuals announce its credibility and community reception:
• GitHub Trending badge declaring “#1 Repository Of The Day.”
• TrendShift badge linking to the project’s presence inTrendShift’s repository rankings.
• Core tagline: an AI-driven search tool scored by upvotes, likes, and real money rather than editorial judgments, delivering a single brief that synthesizes what actually matters in the last 30 days.
• The README positions v3 as the current runtime skill with a dedicated skill spec file (skills/last30days/SKILL.md) that acts as the source of truth for commands and setup behavior.
• The tool boasts zero-configuration operation across major platforms—Reddit, Hacker News, Polymarket, and GitHub—while offering rapid unlocks to YouTube, TikTok, and other sources after a quick setup.

1. Core Idea, Value, and How It Stands Out

• The system aggregates signals from a broad spectrum of public data sources—Reddit, X (Twitter), YouTube, TikTok, Instagram Reels, Hacker News, Polymarket, GitHub, Threads, Pinterest, Bluesky, Perplexity Sonar, and the broader Web—synthesizing them into a single, concise brief.
• It emphasizes “real people engagement” as the currency: upvotes, likes, comments, transcripts, and odds-based signals backed by real-money markets—rather than editorial picks or paid placements.
• The engine’s design is anchored in the philosophy that no single AI has native, equal access to all these ecosystems; the solution is an agent that can bridge multiple platforms with user-provided keys and browser sessions to search in parallel and compare signals directly.
• The ultimate output is a brief that emphasizes what actually matters to people in the last 30 days, with source citations anchored to the original posts or data points.

1. Sources and Signals: What Feeds the Brief

• Primary nodes of data include:
• Reddit: unfiltered discussions with top comments and upvotes delivered publicly.
• X / Twitter: hot takes, expert threads, and rapid reactions.
• YouTube: long-form transcripts and quotable moments extracted from full videos.
• TikTok and Instagram Reels: creator perspectives with captions and engagement signals.
• Hacker News: developer-centric debates and consensus from tech communities.
• Polymarket: odds and market sentiment backed by real money.
• GitHub: people signals (PR velocity, top repos, release notes) and topic signals (issues, discussions).
• Threads, Pinterest, Bluesky: social content and visual signals.
• Perplexity Sonar: grounded web search with citations.
• Web: editorial coverage, blog comparisons, and broader context.
• The README emphasizes the value of cross-source signals: “a Reddit thread with 1,500 upvotes is stronger than a blog post nobody read,” and a TikTok with millions of views can reveal cultural relevance beyond traditional outlets.
• Community contributions continually broaden coverage, with additional platforms like Truth Social and Xiaohongshu (RED) noted as increasingly integrated signals.
• The system merges signals when the same story appears across multiple sources, reducing fragmentation and presenting a unified cluster for the user.

1. Practical Uses and Everyday Scenarios

• Before a meeting: The tool reveals what a person is doing in the current month (for example, Peter Steinberger’s activities in AI teams, PRs, and projects) rather than relying on outdated LinkedIn summaries.
• When something drops: It tracks timely responses across platforms (e.g., Kanye West news and related cultural signals, sponsorship changes, chart debuts) and surfaces what the public is actively reacting to.
• Tool comparison: It facilitates quick, fair comparisons across OpenClaw, Hermes, and Paperclip by presenting side-by-side signals, architecture snapshots, memory models, and security considerations.
• Global understanding: It aggregates signals to provide insights on geopolitical topics, policy events, or market-moving moments.
• Travel and planning: It informs practical, time-sensitive details such as ride closures, crowd sentiment, and community chatter for events or experiences.
• Fast learning: The system demonstrates how JSON-structured prompts and nested formats can prevent “concept bleeding,” favoring prompt reuse and production-quality outputs.

1. What’s New in v3: The Intelligent, Integrated Core

• Intelligent search as the killer feature:
• The engine first determines where to search, resolving relevant handles, channels, and communities before issuing any API calls.
• It uses a new pre-research brain to map topics to the right sources, handles, and subreddits or channels, enabling content discovery that v2 could not reach.
• Examples include how OpenClaw, Paperclip, and Peter Steinberger are resolved to their proper online representations and discussions.
• Best Takes: a built-in qualitative judge scores humor, wit, and virality alongside relevance, ensuring the final brief ends with a “Best Takes” section featuring clever one-liners or viral quotes.
• Cross-source cluster merging: identical narratives across Reddit, X, and YouTube are merged into a single cluster, using entity-based overlap rather than relying on identical titles.
• Single-pass comparisons: entity-aware subqueries enable parallel comparison, reducing multi-pass runtime and delivering results faster.
• GitHub person-mode: topic reasoning pivots to author-scoped queries for people, surfacing what they shipped, where it landed, and corresponding release notes.
• ELI5 mode: a toggle to rewrite synthesis in plain language while preserving data, sources, and citations.
• Expanded data sources: public Reddit comments, YouTube transcripts, Threads, Pinterest, YouTube comments, and more are available or configurable via sources.
• Polymarket enhancements: better disambiguation to separate “Apple” the company from “Apple” the fruit, and a focus on odds rather than dollar volumes.
• Resilience and quality: improved Reddit reliability, timeout budgets, and runtime fallback to ensure robust performance.
• Fun and human-like synthesis: a “Fun judge” scores and weaves clever, shareable humor into the narrative.
• OpenClaw integration: treated as a first-class engine component, with pre-research and frictionless integration for sign-in and data access.
• Testing and reliability: the project touts over a thousand tests and a robust architecture, underscoring a mature, production-ready pipeline.

1. Installation and Setup: How to Bring It Online

• Claude Code route:
• Install: “/plugin marketplace add mvanhorn/last30days-skill”
• Update: “claude plugin update last30days@last30days-skill”
• OpenClaw route:
• Install: “clawhub install last30days-official”
• Gemini CLI route:
• Gemini CLI supports installing extensions from GitHub, but note an upstream installer bug for certain configurations. Workarounds include:
  • Cloning locally and installing from the local path.
  • If the GitHub install fails, use OpenClaw or Claude Code install paths.
• Manual installation:
• “git clone https://github.com/mvanhorn/last30days-skill.git ~/.claude/skills/last30days”
• Zero-config operation: Reddit, HN, Polymarket, and GitHub function immediately after installation; running the tool once unlocks additional sources within roughly 30 seconds.

1. Bring Your Own Keys: Access and Costs

• The system’s multisource capability is built on user-supplied keys and browser sessions, enabling a cross-platform search without inherent cross-platform trust.
• Source-by-source requirements and typical costs:
• Reddit (with comments) + Hacker News + Polymarket + GitHub: no paid keys required; free to use.
• X / Twitter: login in a browser; free to use in normal view.
• YouTube: tools like yt-dlp can be used locally; free to use.
• Bluesky: app password from bsky.app; free access.
• TikTok + Instagram + Threads + Pinterest + YouTube comments: requires ScrapeCreators keys; 10,000 free calls are highlighted.
• Perplexity Sonar: requires an OpenRouter API key; pay-as-you-go model.
• Web search: Brave Search key; with 2,000 free queries per month.
• The approach emphasizes cost-conscious choices, balancing free access with paid keys only where necessary for coverage expansion.

1. How It Works: The Step-by-Step Flow

• Step 1: You input a topic or query (person, company, product, technology, or comparison).
• Step 2: The agent identifies who matters: founders, GitHub repos, subreddits, TikTok hashtags, YouTube channels, and more—returning a map of relevant handles and sources.
• Step 3: All sources are searched in parallel using multi-query expansion, scoring results by engagement, relevance, and freshness.
• Step 4: The engine retrieves depth of data: full YouTube transcripts, top Reddit comments with votes, TikTok captions, Polymarket odds, and more—beyond simple titles and links.
• Step 5: Results are merged into a single story: cross-source aggregation creates a unified narrative cluster, avoiding duplicate items across platforms.
• Step 6: The synthesis is produced as one concise brief, grounded in specific data and clearly cited to original sources, emphasizing what people actually engaged with.
• Step 7: The system becomes your expert: after a single run, the Claude session retains the knowledge, enabling follow-up prompts, drafts, planning, and architecture design grounded in current signals.

1. What People Say: Reactions and Endorsements

• User testimonials highlight how the skill reduces manual research workload, replacing tab-by-tab exploration with a single, coherent brief.
• The perspective is that this approach captures real-time conversations, not just static, outdated blog posts, over the last 30 days.
• The emphasis is on speed, accuracy, and practical utility for decision-making, sales calls, and trip planning.

1. Open Source, License, and Technical Foundation

• The project is distributed under an MIT license and emphasizes transparency:
• No tracking or analytics beyond user-owned data; research stays on your machine.
• The stack includes Python 3.12+, yt-dlp, Node.js (vendored for X search), and ScrapeCreators API.
• The v3 engine architecture is attributed to a contributor, j-sperling, with changelog history maintained in CHANGELOG.md.
• The project celebrates a large test suite (1,012 tests) and a robust, production-ready architecture designed for stability and reliability.

1. Visual and Historical Signals: Star History and Community Momentum

• A Star History graphic is included to visualize the project’s trajectory on GitHub, reflecting community interest and ongoing contributions.
• The repository page and star metrics are presented as part of the visual narrative, reinforcing the project’s open-source status and momentum.

1. Practical Visual Aids for Context

• In-line visuals from the Input:
• The GitHub Trending badge at the top signals community recognition and momentum.
• The TrendShift badge provides external validation of repository status.
• A Star History image shows ongoing engagement and growth over time.
• These visuals accompany the textual description to ground the description in observable signals and community validation.

1. Summary: The Product in One Line

• last30days-skill is a zero-config, AI agent-driven, cross-source search engine that prioritizes what real people have engaged with in the last 30 days, delivering a single, well-sourced brief that blends Reddit, X, YouTube, TikTok, Polymarket, GitHub, and other sources into a coherent narrative, with human-like judgment elements (humor, virality) and robust, source-grounded synthesis.

1. Special Notes on the Experience and Implementation

• The system bridges otherwise isolated ecosystems by letting users bring their own keys and sessions, creating a capable, unified research workflow that previously required tabbed, platform-specific digging.
• It emphasizes speed (single-pass processing) without sacrificing depth (full transcripts, top comments, and market signals).
• The design philosophy centers on making the user’s decisions faster and more informed, rather than simply delivering raw data.

1. Visual Acknowledgments and Licensing Reminders

• The included visuals (GitHub Trending badge, TrendShift badge, and Star History) are part of the original input and serve as visual anchors for credibility and community interest.
• The MIT license governs the project, reinforcing an open, collaborative approach with an emphasis on reproducibility and user control over data.

1. Final Note: What You Get When You Try It

• A ready-to-run skill that requires minimal setup, immediately taps Reddit, HN, Polymarket, and GitHub, and rapidly unlocks additional sources for broader coverage.
• A unifying brief that merges signals from many communities, with a transparent, source-cited narrative focused on what people actually engaged with in the last 30 days.
• A platform-bridging tool designed to accelerate preparation for meetings, trips, and fast-paced decision-making in a world where information arrives from many directions every day.

Images from the Input included in this description:

• 
• 
• 

[![Featured on TechLogHub](https://techloghub.com/badge-neutral.svg)](https://techloghub.com/open-source/vision-claw?utm_source=techloghub_badge&utm_medium=referral&utm_campaign=readme)