AI45Lab/iDeer

What it solves

Manually checking GitHub, arXiv, HuggingFace, and Twitter every day takes roughly 30 minutes. iDeer turns that repetitive patrol into a passive, 5-minute email read. It gathers updates from many platforms, uses an LLM to filter, score, and summarize them, then delivers only what is worth your attention to your inbox at a time you choose.

How it works

You describe your interests in a profile (text description and optionally a researcher profile with Google Scholar links). iDeer then runs a pipeline: it fetches raw items from configured sources, uses an LLM to score and filter them, produces a daily summary per source, and optionally builds a cross-source report and research ideas. Outputs are stored under history/ and can be emailed, shown in a Web UI, or delivered through Telegram/Feishu bots. The system supports scheduled runs (daily, weekdays only, weekly, or monthly) via a built-in scheduler or system cron.

The source layer is plugin-based: new sources can be added by subclassing BaseSource and registering in SOURCE_REGISTRY. Sources include GitHub trending, HuggingFace papers/models, arXiv, PubMed (36M+ biomedical records), Semantic Scholar (200M+ cross-disciplinary papers), X/Twitter, and any RSS feed.

Who it’s for

• AI researchers tracking arXiv papers, new repos, and models, with cross-source association and research idea generation.
• Financial analysts who want periodic industry event summaries (currently under construction).
• Legal scholars and other researchers who need to monitor journal articles across fields via Semantic Scholar, getting relevant papers pushed to email.
• Anyone who wants daily updates without manually browsing multiple platforms.

Highlights

• Multi-source coverage: GitHub, HuggingFace, arXiv, PubMed, Semantic Scholar, X/Twitter, and RSS, with per-source filters (language, categories, search terms, etc.)
• LLM-based curation: scoring, ranking, summarization, and cross-source report generation
• Flexible delivery: email with configurable frequency, Telegram/Feishu bots with command-driven interaction, Web UI with live logs
• Agent-ready skills: provides Claude Code, Codex, and InternShannon skill packages; supports both API-pipeline and chatbot-first modes where agents do their own summarization
• Plugin architecture: easy to add new data sources
• Web UI and desktop client: runs in browser with scheduled admin panel, plus a local GUI option
• Zotero integration: swipe to save papers, sync high-scoring recommendations daily