RepoPulse

English | 简体中文

Discover new and rising GitHub repositories through early traction, growth, freshness, and activity.

RepoPulse is an open-source repository discovery dashboard. It tracks public repository snapshots over time so developers can compare daily, weekly, and monthly momentum while reducing the advantage of old mega-repositories.

What it includes

• Daily, weekly, and monthly discovery views
• Star growth, fork growth, early velocity, freshness, and explainable discovery sorting
• Repository search, language filters, and topic filters
• Responsive desktop and mobile layouts
• URL-backed filters for shareable views
• Supabase schema with RLS and indexed snapshot queries
• GitHub Actions collection every six hours with a broader new-project candidate pool
• A clearly labeled sample-data fallback before Supabase is configured

Architecture

GitHub Actions + short-lived GITHUB_TOKEN
                ↓
         GitHub public REST API
                ↓
     Supabase repositories + snapshots
                ↓
      Next.js Server Components
                ↓
             Vercel

The browser never receives a GitHub token or Supabase service-role key. See docs/architecture.md for the security and data-flow details.

Local development

Requirements:

• Node.js 24
• npm

npm install
npm run dev

Open http://localhost:3000. Without Supabase environment variables, the application intentionally renders a labeled sample dataset so UI work and deployments remain testable.

Supabase setup

1. Create a Supabase project.
2. Apply supabase/migrations/202607160001_initial_schema.sql.
3. Copy .env.example to .env.local and provide the public project URL and anon key.
4. Add SUPABASE_URL and SUPABASE_SERVICE_ROLE_KEY as GitHub Actions repository secrets.

Vercel environment variables:

NEXT_PUBLIC_SUPABASE_URL
NEXT_PUBLIC_SUPABASE_ANON_KEY

GitHub Actions secrets:

SUPABASE_URL
SUPABASE_SERVICE_ROLE_KEY

GITHUB_TOKEN is generated automatically for each Actions job. Do not create a personal token for the MVP collector.

Data collection

Run the collector locally only after setting the required environment variables:

npm run collect

The collector discovers a bounded candidate pool through GitHub Search, deduplicates repositories, scores candidates for discovery value, performs batch upserts, and records one snapshot per repository per hour. It favors newly created repositories, recent pushes, early star velocity, and emerging AI/developer-tool topics instead of simply keeping the largest repositories. The scheduled workflow runs every six hours and safely skips collection until Supabase secrets exist.

AI model

RepoPulse does not need an AI model for the core ranking flow. GitHub Search, repository metadata, Supabase snapshots, and the Postgres ranking function are enough to discover and rank repositories.

The collector can optionally read repository metadata plus a README excerpt for the top 25 discovery candidates and write bilingual structured insights back to Supabase. SenseNova Token Plan uses sensenova-6.7-flash-lite as the primary model. Timeouts, rate limits, and server errors can retry once with deepseek-v4-flash. Existing insights refresh only after 72 hours, and model keys remain server-only.

AI_PROJECT_INSIGHTS_ENABLED
AI_PROJECT_INSIGHTS_LIMIT
SENSENOVA_API_KEY
SENSENOVA_BASE_URL
SENSENOVA_MODEL
DEEPSEEK_API_KEY
DEEPSEEK_BASE_URL
DEEPSEEK_MODEL

Ranking caveat

GitHub does not provide an API field for arbitrary “stars gained in the last seven days.” RepoPulse calculates growth from its own snapshots. Until enough history exists, the site must describe rankings as covering tracked repositories and the period available since tracking began.

See docs/methodology.md for the first ranking formula and known biases.

Commands

npm run dev
npm run lint
npm run typecheck
npm run build
npm run check
npm run collect

Deployment

Deploy the Next.js application to Vercel and connect the main branch. The frontend reads only public ranking data from Supabase. Collection remains isolated in GitHub Actions.

License

MIT