TrendRadar

TrendRadar is an open-source trend and public-opinion monitor from sansan. It collects trending topics from dozens of Chinese platforms plus RSS/Atom feeds, filters them against your interests, and delivers digests to messaging and notification channels. The project is licensed under GPL-3.0, and the homepage lists the latest version as v6.11.0.

What It Is

TrendRadar is a self-hosted news and hot-topic aggregation assistant. It gathers data through the NewsNow API (platforms such as Zhihu, Douyin, bilibili, Weibo and Baidu) and from custom RSS/Atom feeds with freshness filtering. Matching items are ranked and turned into HTML reports and push messages. The core pipeline works without AI; AI features are an optional layer that needs your own provider API key.

Filtering and analysis workflow

Content can be filtered with keyword syntax: required words (+), excluded words (!), /regex/ patterns, @count limits, global filters and display names. Alternatively, AI filtering lets you describe interests in natural language in an ai_interests file, and the AI classifies and scores items against a minimum score, falling back to keyword matching if AI filtering fails. Optional AI analysis covers trend overviews, sentiment, and cross-platform connections, and AI translation can render pushes in other languages. Models are accessed through LiteLLM, which the project says supports 100+ providers including DeepSeek, OpenAI, Gemini and Ollama.

Delivery and scheduling

There are three push modes: daily digest, current rankings, and incremental monitoring. A timeline.yaml scheduler sets collection, push and analysis behavior per time slot, with presets and weekday/weekend differences. Messages can go to WeCom, personal WeChat, Feishu, DingTalk, Telegram, email, ntfy, Bark, Slack and generic webhooks. HTML reports include dark mode, search and keyboard shortcuts, and a web-based visual config editor avoids hand-editing YAML.

Deployment and MCP

TrendRadar can run with Docker (recommended on the homepage), GitHub Actions from a template repository (which requires periodic check-in), or locally with uv on Windows, macOS and Linux. Data is stored in SQLite locally or S3-compatible remote storage such as Cloudflare R2. A separate MCP server exposes analysis tools so MCP-capable clients such as Claude and Cursor can query trending data in natural language.