The Grand Archive

The Grand Archive is a tool that gives your AI agent real-time news search across 20+ sources.

中文版

---

Quick Start — For AI Agent Users

Add The Grand Archive skills to your Claude project in one command:

npx skills add LittleLittleCloud/The-Grand-Archive

Then tell your Claude: "Search recent news about tariffs" — that's it.

Your AI agent can now search, summarize, and analyze 14,000+ live news entries from Bloomberg, AP News, Hacker News, and more.

---

Quick Start — For Developers

Install the TypeScript SDK:

npm install @littlelittlecloud/dak

import { DakClient } from "@littlelittlecloud/dak";

const client = new DakClient({ baseUrl: "https://dak-news.com" });

const result = await client.search({ q: "tariff", category: "finance", limit: 10 });
console.log(result.results[0].title);

Or use the CLI:

npm install -g @littlelittlecloud/dak-cli
dak search "AI regulation" --category tech

MCP endpoint (curl + bash)

The server now exposes an MCP endpoint with the dak_search tool and OAuth metadata:

• MCP JSON-RPC endpoint: POST https://dak-news.com/mcp
• OAuth authorization endpoint (Authorization Code + PKCE): GET https://dak-news.com/oauth/authorize
• OAuth metadata:
  • GET https://dak-news.com/.well-known/oauth-protected-resource
  • GET https://dak-news.com/.well-known/oauth-authorization-server
• OAuth dynamic client registration (public PKCE clients): POST https://dak-news.com/oauth/register

MCP tools include:

• dak_search
• dak_digest_create
• dak_digest_list
• dak_digest_edit
• dak_digest_delete
• dak_api_key_create

Example (list tools):

curl -sS https://dak-news.com/mcp \
  -H 'content-type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}' | jq

Example (dak_search call):

curl -sS https://dak-news.com/mcp \
  -H 'content-type: application/json' \
  -H "authorization: ******" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"dak_search","arguments":{"q":"tariff","category":"finance","limit":5}}}' | jq

---

Use Cases

• "Summarize today's finance headlines" — get a structured digest from Bloomberg, CNBC, MarketWatch, and more
• "Track cross-source coverage of a topic" — compare how AP News, Al Jazeera, and BBC cover the same event
• "Deep-dive analysis with citations" — pull 30+ articles on a topic, cluster by theme, output a research brief

---

Features

• AI-native — Claude Skills included; your agent searches, filters, and summarizes without extra setup
• 21 live sources — finance, geopolitics, tech, and social trending, updated every 30 minutes
• Full-text search — fuzzy + prefix matching across 14,000+ entries via MiniSearch
• TypeScript SDK — typed HTTP client, supports ESM and CJS
• CLI — dak command for one-liner searches and feed browsing
• Tiered access — anonymous (28 days), free (90 days), premium (unlimited)

---

Sources

Category	Sources
International / Geopolitics	BBC Chinese · NYT Chinese · Al Jazeera · AP News · Foreign Affairs · The Diplomat · 参考消息 · 人民网
Finance / Macro	Bloomberg · CNBC · MarketWatch · 华尔街见闻 · 第一财经 · 财新网 · ZeroHedge · 金十数据 · 雪球
Social Trending	Weibo Hot · Zhihu Hot
Tech	Hacker News

Full configuration: config/sources.yaml

---

Skills (AI Agent Integration)

Two Claude Code / AI Agent skills are included in the skills/ directory.

skills/dak — Search & Browse

Search and access feed data from The Grand Archive

Supports full-text search, filtering by category / source / tag / date range.

> Search for finance news about tariffs
> Find recent articles about AI with dak
> Look up oil price reports from March 2026

Trigger keywords: search, find articles, dak search, dak feeds

skills/dak_summary — Structured Analysis

Methodology for structured analysis built on feed data

Three-step core: Broad Query → Cross-Source Analysis → Structured Output

1. Broad Query — retrieve widely from 14,000+ entries using multiple keyword passes
2. Cross-Source Analysis — dedup, cross-validate, identify patterns (timeline / causation / trends), cluster by theme
3. Structured Output — write to a structured summary file

Two key use cases:

• Daily digest — run analysis over all feeds for a given day, output a topic-grouped summary
• Topic deep-dive — trace a topic across time, produce a chronological event timeline

> Summarize today's news
> Analyze how the Iran conflict affects oil and gold prices
> Write a topic summary on recent AI developments

Trigger keywords: summarize, daily digest, analyze [event], write topic summary

Install Skills

Recommended — one command:

npx skills add LittleLittleCloud/The-Grand-Archive

Or copy manually:

cp -r skills/dak /your-project/.claude/skills/dak
cp -r skills/dak_summary /your-project/.claude/skills/dak_summary
npm install -g @littlelittlecloud/dak-cli

---

CLI Reference

The CLI connects to https://dak-news.com by default. Set an API key to unlock premium access:

export DAK_API_KEY=your-api-key  # required for premium tier (90-day history and higher rate limits)

Search

dak search "inflation"
dak search "oil price" --category finance --from 2026-03-01
dak search "AI" --limit 10
dak search "tariff" --json

Browse

dak feeds
dak feeds --category tech --limit 20
dak feeds --source Bloomberg --limit 10

Other

dak stats
dak suggest "inflat"

Options

Option	Description
--category <cat>	Filter by category: finance / news / social / tech / blog / podcast
--source <src>	Filter by source
--from <date>	Published after (ISO 8601)
--to <date>	Published before (ISO 8601)
--limit <n>	Max results (default 20)
--json	JSON output

---

SDK Reference

import { DakClient } from "@littlelittlecloud/dak";

const client = new DakClient({
  baseUrl: "https://dak-news.com",
  apiKey: "your-api-key", // optional
});

// Search
const result = await client.search({
  q: "inflation",
  category: "finance",
  from: "2026-03-01",
  limit: 20,
});
result.results[0].title;  // article title
result.results[0].score;  // relevance score
result.total;             // total matches
result.tier;              // "anonymous" | "free" | "premium"

// Browse
const feeds = await client.getFeeds({ category: "tech", limit: 10 });
const finance = await client.getFeedsByCategory("finance");
const bloomberg = await client.getFeedsBySource("Bloomberg");

// Single entry
const entry = await client.getFeed("entry-id");

// Stats
const stats = await client.getStats();
stats.total;        // total entry count
stats.byCategory;   // [{ category, count }]
stats.bySource;     // [{ source, count }]

API Methods

Method	Description
search(params)	Full-text search with category / source / date / pagination filters
getFeeds(params?)	Browse entries with multi-dimensional filters
getFeed(id)	Get single entry by ID
getFeedsByCategory(cat)	Get entries by category
getFeedsBySource(src)	Get entries by source
getStats()	Get index statistics

---

Technical Details

Search Capabilities

Built on MiniSearch — full-text search across 14,000+ entries:

• Title search — fast matching (364ms index build, low memory)
• Fuzzy matching — ~20% character distance tolerance (inflaton → inflation)
• Prefix matching — inflat matches inflation, inflationary
• Multi-dimensional filters — category, source, date range
• Tier limits — anonymous (28 days), free (90 days), premium (unlimited)

Architecture

┌─────────────┐    ┌─────────────────┐    ┌──────────────┐
│  RSS Feeds  │───▶│ Ingestion Worker│───▶│   SQLite DB  │
└─────────────┘    └─────────────────┘    └──────┬───────┘
                                                 │
                          ┌──────────────────────┤
                          ▼                      ▼
                   ┌─────────────┐       ┌──────────────┐
                   │  Hono API   │       │  MiniSearch  │
                   │   Server    │       │    Index     │
                   └──────┬──────┘       └──────────────┘
                          │
            ┌─────────────┼─────────────┐
            ▼             ▼             ▼
      ┌──────────┐ ┌──────────┐ ┌──────────┐
      │   SDK    │ │   CLI    │ │   Web UI │
      └──────────┘ └──────────┘ └──────────┘

npm Packages

Package	Description
@littlelittlecloud/dak	TypeScript SDK — HTTP client, ESM / CJS
@littlelittlecloud/dak-cli	CLI tool — dak command, zero-dependency single-file bundle

Project Structure

packages/
  contract/          # Zod schemas, types, route constants (workspace-only)
  server/            # Hono API server + SQLite + MiniSearch
  sdk/               # @littlelittlecloud/dak — HTTP client SDK
  cli/               # @littlelittlecloud/dak-cli — CLI tool
  ingestion-worker/  # RSS fetch worker (every 30 minutes)
  ui/                # Web dashboard

Local Development

bun install

# Start API server
cd packages/server && bun run dev

# Run ingestion worker (one-shot mode)
cd packages/ingestion-worker && ONCE=1 bun run src/index.ts

# CLI development
cd packages/cli && bun run src/index.ts search "test"

RSS Feed Collection

bun run fetch              # fetch all enabled sources
bun run fetch finance      # fetch only finance category
bun run list               # list saved entries
bun run sources            # view all sources

Adding a Source

Edit config/sources.yaml:

- name: Source Name
  url: "{{RSSHUB_BASE_URL}}/example/route"
  category: tech
  enabled: true
  fetch_interval: 3600
  tags: [tag1, tag2]
  description: Short description

Building npm Packages

cd packages/sdk && bun run build
cd packages/cli && bun run build

Deployment

Server and Worker are deployed on Fly.io (Tokyo nrt):

bun run deploy:server
cd packages/ingestion-worker && fly deploy

bun run deploy:server runs a pre-deploy guard that fails unless dak-server has exactly one Fly machine and one attached dak_data volume.

Dashboard

bun run dev     # development server
bun run build   # production build
bun run deploy  # deploy to GitHub Pages

CI/CD

Workflow	Trigger	Action
fetch-feeds	Every 30 min / manual	Fetch RSS → commit new feed files
publish-dak	After feeds update / manual	Build and publish @littlelittlecloud/dak + @littlelittlecloud/dak-cli to npm
deploy	Manual	Build and deploy Dashboard to GitHub Pages

---

License

MIT