Find GitHub repos before they trend.
See the repos gaining stars fastest right now, filter them in seconds and keep a watchlist.

Try it live · See it in action · Highlights · More by juw12

Movers and rising Repos gaining the most stars in the last 24 hours, ranked.

Filter fast Narrow the list by language, stars and age.

Watchlist Keep an eye on repos before everyone else does.

See it in action

_{Open the interactive demo at juw12.github.io/earlystar/}

About

Find GitHub repos before they trend. Live at juw12.github.io/earlystar.

EarlyStar tracks GitHub Trending and fast-rising new repositories, keeps its own star history, and ranks repos by the stars they actually gained.

• Movers: biggest real star gains in 24 h / 7 d / 30 d, with the source of every number shown (tracked by EarlyStar, a brand-new repo, GitHub Trending's own count, or an estimate).
• Rising: brand-new repos ranked by stars per day and a momentum score.
• Trending log: a day-by-day archive of GitHub Trending with streaks and appearance counts.
• Repo details: star-history chart, Trending appearances, language breakdown, README, and exact 24 h / 7 d gains from stargazer timestamps (with your own GitHub token).
• Activity, not just stars: commits, merged PRs, new issues (last 7 days) and the latest release for the hottest repos, so hype can be told apart from a project that's actually shipping.
• README badges + open JSON API: every tracked repo gets a live badge and all data is plain JSON (see below).
• Discovery tools: "new to me" and "hide seen", a watchlist, side-by-side compare, CSV export, keyboard shortcuts, light and dark themes.

Apps

Desktop (Windows & macOS)

Prefer a real app window? Download it (about 5–10 MB; it opens the live site, so data is always current):

Platform	Download
Windows 10/11	EarlyStar-Setup-x64.exe · portable .exe
macOS 10.15+ (Apple Silicon + Intel)	EarlyStar-macOS.dmg

The apps aren't code-signed yet. On Windows, if SmartScreen appears click More info → Run anyway. On macOS, drag EarlyStar to Applications, then right-click it → Open the first time.

Phone (iPhone & Android)

No app store needed: EarlyStar installs straight from the website as an app with its own icon, full screen.

• iPhone / iPad: open juw12.github.io/earlystar in Safari → Share button → Add to Home Screen.
• Android: open it in Chrome → tap Install app (or ⋮ menu → Install app / Add to Home screen).

Badge for your README

Open any repo on EarlyStar and copy the snippet from its details panel, or build it yourself:

[![EarlyStar](https://img.shields.io/endpoint?url=https%3A%2F%2Fjuw12.github.io%2Fearlystar%2Fdata%2Fbadge%2FOWNER%2FNAME.json)](https://juw12.github.io/earlystar/#repo=OWNER/NAME)

It shows #3 today · +3.8k★ while the repo is a top-100 mover, +X★ this week otherwise, and falls back to the total stars.

JSON API

Static files on GitHub Pages, refreshed every 30 minutes, CORS-friendly, no key needed. Base URL: https://juw12.github.io/earlystar/data/

File	What's in it
health.json	Last collection time and counts
movers-24h.json, movers-7d.json, movers-30d.json	Ranked star gains (items[] with gain, gain_source, engagement, …)
trending/days.json	Days with a recorded GitHub Trending snapshot
trending/YYYY-MM-DD.json	That day's Trending lists, keyed since|language
repos.json	owner/name → repo id
history/{id}.json	Star/fork snapshots, Trending appearances and activity for one repo
badge/{owner}/{name}.json	shields.io endpoint payload

How it works

A GitHub Action (.github/workflows/collect.yml) runs every 30 minutes (at :07 and :37, away from the busy top of the hour):

1. collector.py scrapes GitHub Trending and runs a few search queries, saving star snapshots into SQLite, then fetches 7-day activity for the hottest repos via one batched GraphQL query.
2. The database is kept on the data branch (one commit, overwritten each run, so history doesn't bloat the repo).
3. export_static.py turns the history into JSON files and badges, and the site is deployed to GitHub Pages.

index.html is a single file with no build step. It reads those JSON files on GitHub Pages, talks to server.py when run locally, and still works opened straight from disk (Rising / Established / Watchlist only).

Calls: play-money predictions on repos

Spot a repo early, then call it: "Will owner/repo reach 10k★ by Dec 31?", "Will A out-star B this week?", "Will it trend 4 of the next 7 days?" Calls settle themselves from EarlyStar's own snapshots, so there's no human judge and nothing to dispute.

Points are play money. They have no cash value and can't be bought, sold, transferred or withdrawn.

What you get

• Calls list with live odds, 24 h change, volume, traders and progress toward the target.
• Market drawer: odds chart, stars-vs-target chart (or a versus/Trending view), trade panel with a live quote, your position and P&L, the settlement rules, recent trades and top holders.
• Trading against an LMSR automated market maker, so every call has a price from the first second. Slippage protection is built in.
• New call: pick a repo and EarlyStar suggests a target and opening odds from its momentum model. Creating a call costs 50 points.
• Auto calls generated from the radar every 6 hours: star targets for top movers, Trending streaks, and versus pairs.
• Portfolio, Leaderboard (profit, accuracy, best early calls) and public profiles.
• Radar integration: every repo card shows its live odds and has a "Call it" button.
• Integrity: owners can't trade their own repos, accounts must be at least 14 days old, at most 500 points at risk per call, rate limits, and a star-spike review before paying out a suspicious YES.

How calls settle

Call	Settles YES when…	Otherwise
Star target	any snapshot taken before the deadline shows ≥ target (early)	uses the first snapshot taken at or after the deadline. If there is none within 6 h, it falls back to the last snapshot in the 3 h before the deadline; if that's missing too, the call is voided and everyone is refunded
Versus	A gained strictly more stars than B since the call opened	a tie or missing data voids the call
Trending	the repo is logged on daily Trending (all languages) on N distinct days (early)	NO at the end of the window. If EarlyStar missed days and those days could have changed the answer, the call is voided

A YES that follows a star spike (≥ 200★ in 48 h and more than 5× the prior week's rate) goes to review, and an admin settles or voids it.

Run it locally

python3 server.py --dev-login --open     # http://127.0.0.1:8765

• --dev-login lets you sign in with any username so you can try everything locally. It is refused together with --public.
• The first collection takes 1–2 minutes. Auto calls appear once fresh data exists.
• Optional: export GITHUB_TOKEN=ghp_... (a token with no scopes is fine) for higher GitHub rate limits.
• Make yourself an admin: export EARLYSTAR_ADMINS=juw12. Admins see the review queue and can settle or void calls.
• On Windows, double-click start.bat. On macOS/Linux, run ./start.sh.

Needs Python 3.9 or newer. It uses only the standard library, so there's nothing to install.

Deploy with accounts (public server)

1. Create a GitHub OAuth app at https://github.com/settings/developers → New OAuth App:

   • Homepage URL: https://your-domain
   • Authorization callback URL: https://your-domain/auth/callback

2. Set these environment variables:

   Variable	Value
   EARLYSTAR_GITHUB_CLIENT_ID / EARLYSTAR_GITHUB_CLIENT_SECRET	from the OAuth app
   EARLYSTAR_BASE_URL	https://your-domain (https turns on Secure cookies)
   EARLYSTAR_ADMINS	comma-separated GitHub logins
   GITHUB_TOKEN	a token with no scopes, for the collector's rate limits
   EARLYSTAR_DATA_DIR	a persistent volume path, e.g. /data

3. Run python3 server.py --public (binds 0.0.0.0:$PORT), or use the Dockerfile:

   docker build -t earlystar .
   docker run -p 8765:8765 -v earlystar-data:/data --env-file .env earlystar

   This works on Fly.io, Render, Railway or any small VPS. Use a persistent volume. The app database holds every account and balance.

--public restricts the old single-user endpoints (/api/token, /api/collect, POST /api/watch) to admins, requires sign-in for /api/stargains, and hides internal error text.

Architecture

collector.py   GitHub Trending + search → data/radar.db (public, rebuildable). Also refreshes repos with open calls
markets.py     schema, LMSR pricing, trading, creation, settlement, auto-generation, scoring
auth.py        GitHub OAuth (no scopes), dev login, hashed cookie sessions
server.py      JSON API, static index, collector thread, market service thread
index.html     single-file UI, no build step
tests/         python3 -m unittest discover -s tests -t .

• Two databases. data/radar.db is public and the GitHub Action pushes it to the data branch for the static Pages mirror. data/earlystar.db holds users, sessions, markets, trades and the ledger, and must stay private. An old data/reporadar.db is renamed to radar.db automatically.
• Ledger invariant. Every balance change is a ledger row, and the tests check that users.balance == SUM(ledger.delta).
• Static mirror. The GitHub Pages site (export_static.py) still serves the read-only radar. Calls need the server.

See PLAN.md for the full design: pricing math, scoring and the security model.

License

Released under the MIT License © 2026 Juwen Lie. Contributions are welcome under the same license; see the contributing guide and Code of Conduct.