TrendRadar: Rebuilding Digital Cognition in the Information Age

> In an era of algorithmic feeding, TrendRadar is a lighthouse in the digital ocean—it not only shows you where the reefs are, but teaches you to steer your own ship. It is both a technical achievement and a quiet revolution in information sovereignty.

Every day, more than 2.5 exabytes of information pour in from around the world. Zhihu threads, Douyin hits, Bilibili comments, Weibo trending topics—platforms shout at us simultaneously, shredding our attention into fragments. Our problem is not information scarcity but cognitive paralysis from information overload. TrendRadar, an open-source project by GitHub user sansan0, takes the opposite path from recommendation algorithms: it hands the power of choice entirely back to the user.

Chapter 1: Minimalist Engineering

TrendRadar's DNA is built on two words: lightweight and easy to deploy. From v1.0.0 (June 16, 2025) to v3.4.1 (November 28, 2025), the project went through at least 15 major iterations, added 8 push channels and an AI analysis module—yet the core main.py was deliberately never split up, so users can copy-and-paste upgrade. The user's upgrade cost is borne by the developer.

Version 2.0.0 (July 17, 2025) centralized all configuration in config/config.yaml and introduced three run modes: daily (day summary), current (current rankings), and incremental (delta monitoring). Version 3.2.0 added keyword ranking control and precise display limits—for example showing 20 items for "Tesla" (@20) but only 5 for "BYD" (@5).

Deployment can be as simple as three steps: Fork the repository, set GitHub Secrets, and run GitHub Actions once.

Chapter 2: Architecture

Data Collection

Data comes from the open-source newsnow multi-platform aggregator API. The author thanks the newsnow maintainer and reminds users the service runs on his goodwill and trust. The default 11 monitored platforms form a sampling matrix:

• Social: Weibo, Douyin, Bilibili, Zhihu, Tieba
• News portals: Toutiao, Baidu Hot Search, Fenghuang, The Paper
• Finance: Wallstreetcn, Cailianshe

The author warns: more platforms is not better—filter, or you will be drowned again.

Keyword Filtering

Four syntax elements form a miniature query language: 1. Normal terms: basic match (e.g., Huawei, Apple) 2. Required terms (+): title must also contain this 3. Filter terms (!): exclude matching items 4. Count limits (@): control displayed items

Blank lines separate independent keyword groups, like casting multiple nets over different fish schools.

Weighted Ranking

A three-weight hybrid model scores items:

• Rank weight (60%): position on platform rankings
• Frequency weight (30%): how often it appears during monitoring
• Heat weight (10%): absolute ranking quality

Tuning suggestions: set rank_weight to 0.8 for real-time trend chasing (self-media creators); raise frequency_weight to 0.5 for deep-topic tracking (investors).

Chapter 3: AI via MCP

Version 3.0.0 (October 20, 2025) integrated AI analysis through the Model Context Protocol (MCP)—think of it as the USB-C of the AI world. The MCP server (as of v1.0.3) exposes 14 tools:

• Basic queries: get_latest_news, get_news_by_date, get_trending_topics
• Smart retrieval: search_news, search_related_news_history
• Analysis: analyze_topic_trend, analyze_data_insights, analyze_sentiment
• System: get_current_config, get_system_status

Users can conversationally chain questions—"query yesterday's Zhihu hot topics," "analyze these trends," "generate a summary report"—with the AI auto-invoking tools into a complete analysis chain. The author notes honestly: AI features require locally accumulated data; the bundled Nov 1–15 test data is only a demo. Data sovereignty stays with the user.

Chapter 4: Push Philosophy

Eight push channels are supported, from WeChat Work and Telegram to email and Slack. Three push modes act like different personalities:

1. Daily mode: a diligent secretary delivering a complete daily report at a fixed time—even items pushed in the morning reappear in the 6 PM summary. 2. Current mode: a live commentator reporting what's hot right now, updating as items rise or fall. 3. Incremental mode: a shrewd intelligence agent reporting only *new* developments—zero noise for investors and traders.

Version 2.4.0 added push time windows (e.g., only 9:00–18:00), respecting users' life boundaries in an always-on culture: not every moment needs to be filled with information.

Chapter 5: Deployment

Deployment follows a progressive-complexity design:

• GitHub Fork (~30 seconds): Fork → set Secrets → run Actions once. The README's 8 collapsible Secrets tutorials include strict Name-field requirements, step-by-step guides, and mobile/PC differences—defensive documentation that pre-fills every pitfall.
• Docker: one-line quick start, recommended docker-compose with env-var override, or local build. Image wantcat/trendradar is multi-architecture (x86 servers to ARM Raspberry Pi).
• v3.0.5 added environment-variable-over-config.yaml priority after NAS users hit config issues—a rapid response to real pain points that helped the project earn 1000+ stars.

Chapter 6: Community

The README's acknowledgment ledger records 73 supporters with individualized notes—from a 1.8-yuan triple-tip from a user with a cat avatar to "supporting Chinese open source!" The author's stance on stars: "Fork is what you desire, star is what I desire; having both is the best support for open-source spirit." The project was featured by Xiaozhuanruan (Small Software), LinuxDo, and Ruanyifeng's weekly newsletter, and is sponsored by 302.AI and LightningTalk, with usage tutorials written for the sponsors.

Chapter 7: Future Directions

Hidden in the version numbers are clear trajectories:

• AI nativization: ongoing MCP refinements (date parsing, parameters, tool count) suggest future auto keyword recommendation, trend prediction, and cross-language analysis.
• Push channel expansion: possibly Discord, WhatsApp, or decentralized protocols like Matrix.
• Data loop closure: custom platform interfaces already exist; future directions include self-hosted sources, intranet monitoring, and vertical-domain mining—evolving TrendRadar from aggregator to information middleware.

Conclusion: A Declaration of Information Sovereignty

TrendRadar's ultimate value is not how many platforms it aggregates or how smart its AI is, but the signal it sends: in an age ruled by algorithms, ordinary people can still reclaim information sovereignty. Like the printing press moving knowledge from monastery scrolls to common paper, TrendRadar moves the power of information selection from black-box algorithms back to every user. Configure a few keywords, pick a push mode, set a channel—and you own a personal information radar, deployable in about 30 seconds.

References

1. sansan0. (2025). *TrendRadar: Hot-topic assistant deployable in 30 seconds*. https://github.com/sansan0/TrendRadar 2. sansan0. (2025). *TrendRadar MCP module changelog*. https://github.com/sansan0/TrendRadar/releases 3. ourongxing. (2025). *newsnow: Multi-platform hot-topic aggregation API*. https://github.com/ourongxing/newsnow 4. Model Context Protocol. (2025). *MCP official documentation*. https://modelcontextprotocol.io/ 5. Ruanyifeng. (2025). *Science and Technology Weekly*. https://github.com/ruanyf/weekly