AI Trend Radar: trending Hugging Face models, datasets, Spaces, papers & GitHub AI repos in one ranked feed

See what's taking off in AI before it shows up in newsletters. AI Trend Radar pulls six live public sources in one run: Hugging Face trending models, datasets, Spaces, Hugging Face daily papers (arXiv), and new GitHub AI repositories. It ranks everything with a single momentum score and flags breakouts. Run it on a schedule and it also tells you what is new since the last run and what moved up, so the output can feed alerts directly.

Most Hugging Face or GitHub-trending scrapers cover one source and give you raw lists. This Actor merges the sources, normalises growth into per-day velocity, and scores items against each other.

What you get

Field	Meaning
overallRank, momentumScore (0-100)	Cross-source ranking. 50% velocity percentile within the source, 50% the source's own trending-rank percentile
signal	breakout / hot / rising / steady
likesPerDay	Likes, upvotes or stars per day since creation (velocity)
isNewSinceLastRun, rankChangeSinceLastRun, metricDeltaSinceLastRun	Change tracking for alerts and dashboards
likes, downloads, stars, trendingScore, license, category, tags, description	Source metadata

Example output (actual top rows from a default run on 2026-10-14)

#	source	name	category	momentumScore	signal	velocity / day	cross-source
1	github_repos	dsh-our-free-model	agents	100.0	hot	795.31	1
2	github_trending	rea	agents	100.0	hot	41187.0	1
3	huggingface_datasets	tiktok-5.6B-videos	vision	99.0	breakout	26.01	1
4	huggingface_spaces	MiniMax-H3-Turbo-Lora-UNCENSORED	audio	99.0	hot	25.45	2
5	huggingface_datasets	MiMo-V2.6-RL-oss	dataset	98.0	hot	48.64	1
6	github_repos	esp32-iphone-agent	agents	98.0	breakout	659.7	1
7	huggingface_spaces	jev-decision-index	other	97.0	hot	24.18	2
8	github_repos	magpie	agents	96.0	hot	597.26	1

That run scored 256 items across 6 sources, found 7 cross-source projects and flagged 8 breakouts. Velocity is likes, upvotes or stars per day.

New in v0.2

• GitHub Trending feed: AI-related repos from github.com/trending with starsGainedInPeriod (daily, weekly or monthly). Repos found by both GitHub feeds are merged into one row.
• Hugging Face Space details: runtime stage, hardware (e.g. zero-a10g), SDK, and the models and datasets each Space uses.
• Category tagging: categories / primaryCategory = llm, vision, audio, agents, code, embeddings-rag, robotics, dataset, research. Filter with categories.
• Cross-source project merge: a paper, its GitHub repo, its model and the Space that demos it share a projectId, with relatedItems, crossSourceCount and a boosted projectMomentumScore. Matching uses explicit links (paper→repo, arXiv tags, Space→models, base models) plus exact normalised-name matches across sources.
• Digest record: the key-value store record DIGEST holds the top 10, breakouts, top 3 per category, category counts, new since last run, biggest rank gainers and cross-source projects. It's ready for a newsletter, a Slack post or a dashboard.

Use cases

• AI newsletters and creators: a ranked "what's hot this week" list in one API call.
• Investors and analysts: spot fast-rising open models, labs and repos early.
• ML engineers: keep track of new models, datasets and papers in your niche with hfSearch (e.g. embedding, tts, qwen).
• Alerts: schedule it hourly or daily with onlyNewSinceLastRun + minMomentumScore: 70, then pipe the results to Slack, email or a webhook through Apify integrations.

Input

Pick sources, items per source, an optional Hugging Face keyword, GitHub topics and window, and filters. An optional GitHub token raises rate limits. See the Input tab.

Pricing

Pay per event: you're charged once per output item (trend-item). The Actor respects your run's spending limit and stops cleanly when it's reached.

Data sources & notes

Data is fetched live from the public Hugging Face Hub API, the GitHub Search API and the public github.com/trending page. Papers with Code has been retired (it now redirects to Hugging Face Papers, which is included). Ollama and LM Arena leaderboards are left out until their reuse terms are clear. Scores are relative rankings computed from those public numbers. They aren't forecasts. Third-party content keeps its original licenses (see license).

More from Intenal_Builds

• Pump.fun Graduated Tokens Dataset - every pump.fun graduation on Solana with market-cap path
• Crypto Launch Radar - new tokens and pairs from DEX Screener and GeckoTerminal
• Memecoin Alpha Caller Leaderboard - pump.fun KOL caller hit rates and consistency