OpenGIKAI — Opening Up Parliament

open-gikai.net | 🇯🇵 日本語版はこちら / Japanese

OpenGIKAI (議会) is an open-source public media project that transforms Japanese parliamentary proceedings into a modern, accessible thread format — like social media, but with official sources. It ingests multiple sources including Diet records (NDL), Prime Minister press conferences (kantei.go.jp), and government council meeting minutes (審議会).

What It Does

• Fetches official transcripts from multiple sources: NDL Diet Records API, kantei.go.jp press conferences, and government council meeting minutes
• Uses AI (Claude) to summarize and structure speeches by topic
• Links each thread to related news articles with image previews (Bing News)
• Presents them in a thread-based UI with three reading levels:
  • 🌱 Easy — Simple language for everyone
  • 📖 Standard — Balanced detail with brief explanations
  • 📰 Detailed — Full political context, news-style

Why

Parliamentary records are public but hard to read. OpenGIKAI makes them accessible without editorializing — every summary links back to the original transcript. The AI prompts and processing logic are fully open-source to ensure transparency and political neutrality.

Tech Stack

Layer	Technology
Frontend	Next.js 16 (App Router), TypeScript, Tailwind CSS
Deployment	SSG frontend: Cloudflare Workers static assets (Workers Builds). MCP server (apps/mcp/): Vercel
Data Pipeline	Python + Claude API (Message Batches API + prompt caching)
Data Sources	NDL Diet Records API, kantei.go.jp, cao.go.jp councils
Public API	Read-only MCP server for Claude Desktop / Cline / custom agents

Getting Started

# Clone the repository
git clone https://github.com/wharfe/open-gikai.git
cd open-gikai

# Install frontend dependencies
npm install

# Start the frontend dev server
npm run dev

The MCP server is a separate Next.js project under apps/mcp/ with its own dependencies:

cd apps/mcp
npm install
npm run dev   # serves on http://localhost:3100

See apps/mcp/README.md for MCP deployment details.

Project Structure

├── src/                  # Frontend (Next.js SSG — output: "export")
│   ├── app/              # App Router pages
│   ├── components/       # React components
│   ├── lib/              # Utilities and data fetching
│   └── types/            # TypeScript type definitions
├── apps/
│   └── mcp/              # MCP server (Vercel project, dynamic Node runtime)
├── scripts/              # Python batch processing
│   ├── sources/          # Source adapters (NDL, kantei, council, …)
│   └── pipeline/         # AI pipeline (grouping, summarization, news ranker)
├── data/                 # Generated JSON consumed by both frontend SSG and MCP server
│   ├── threads/          # Per-date thread files
│   └── members.json      # Accumulated Diet member registry
├── public/               # Static assets (incl. sitemap, RSS feed)
└── .github/workflows/    # daily-batch.yml (6:00 AM JST cron)

The frontend uses output: "export", so anything requiring a Node runtime (Route Handlers, dynamic APIs) lives under apps/.

How It Works

Sources (NDL, kantei, council)
   ├─► fetch (sliding 30-day window per run)
   │
   ├─► group by topic                  ┐
   ├─► classify tension                │  Claude API
   ├─► summarize at 3 levels (Batches) │  + prompt caching
   ├─► extract commitments / outcomes  ┘
   │
   ├─► enrich with related news (Bing News + Claude relevance ranker)
   │
   └─► generate JSON
         ├─► frontend SSG  → open-gikai.net (Cloudflare Workers)
         └─► MCP server     → /api/mcp       (Vercel, apps/mcp)

This is the beginning of the README. Read the rest at the original. Read the full README ↗