DEV Community

heubme
heubme

Posted on

LlamaIndex Reader: DataSinking

#llm #mcp

Load full-text financial reports (balance sheet, income statement, cash flow + notes) as LlamaIndex Documents for your RAG pipeline. Makes DataSinking the first step of "build a financial-report RAG".

Install 

pip install git+https://github.com/heubme2020/llama-index-readers-datasinking
Enter fullscreen mode Exit fullscreen mode

The query-engine example below also needs LlamaIndex itself: pip install llama-index.

Usage (20-line Colab) 

from llama_index.readers.datasinking import DataSinkingReader

# Get a free key at datasink.ing; leave empty for the public quota (31 docs / 7 days / IP)
reader = DataSinkingReader(api_key="your key")
docs = reader.load_data("AAPL", limit=3)   # Apple's 3 latest filings

from llama_index.core import VectorStoreIndex
index = VectorStoreIndex.from_documents(docs)
print(index.as_query_engine().query("What was Apple's net income last year?"))
Enter fullscreen mode Exit fullscreen mode

Each report becomes one Document; its Markdown tables are preserved, so the LLM can read the actual balance sheet / income statement / cash flow.

Three examples

The repo's examples.py has three runnable examples. Set your key as an env var and run:

export DATASINKING_API_KEY=your_free_key    # Windows: set DATASINKING_API_KEY=your_free_key
python examples.py
Enter fullscreen mode Exit fullscreen mode
  1. Fetch one filing — see what a Document looks like (metadata + full text).
  2. Load 5 years of filings — 626k characters of Apple filings as 5 Documents, ready for VectorStoreIndex.from_documents.
  3. Cross-company comparison — search "营业收入" across Moutai vs Wuliangye and read the revenue figures side by side: 
for sym, name in [("600519.SS", "茅台"), ("000858.SZ", "五粮液")]:
    docs = reader.load_data(sym, limit=1)
    # 茅台 → | 营业收入 | 90,703,260,964.48 | ...  ·  五粮液 → | 营业收入（元） | 28,416,674,541.77 | ...
Enter fullscreen mode Exit fullscreen mode

API key, quotas and rate limits

You don't need a key to start, but the public tier is throttled — use a (free) key for real RAG pipelines.

Tier Quota Rate limit
No key (public) 31 reports / 7 days / IP ~1 request / 3 s
Free key 8,191 reports / 7 days 3 requests / s
Paid ($31/yr) 524,287 reports / 7 days 31 requests / s
  • Quota counts reports, not requests — one report = one unit.
  • Hit a limit and you get a QuotaError carrying the tier table above plus the upgrade link (https://datasink.ing/pricing).

Supported symbols

  • US AAPL · China 600519.SS · Japan 7203.T · Korea 005930.KS · Taiwan 2330.TW · UK VOD.L
  • Don't know the ticker? Search by name: GET https://api.datasink.ing/search?q=Apple
  • Each report is a Document with metadata symbol / report_period / doc_type

API docs: https://datasink.ing/docs

Top comments (0)