Agenticx Query Data Source

Query Data Source

When to use

• Any question about stock price, financial indicators, macro economic data, company registry, academic citation counts, or legal statute text that has a real, checkable current value.
• Do NOT answer from training memory for these categories — training data is stale and the user can verify against a real source.
• Call list_data_sources first when unsure which plugin or API fits (optional domain: finance / macro / academic / enterprise / legal).

How to call

1. list_data_sources(domain="finance") — discover enabled plugins and api_name values.
2. query_data_source(data_source_name=..., api_name=..., params={...}) — fetch structured JSON.
   • Default chart window: days: 60 (≈3 个月交易日). A candlestick chart with only 5–10 bars looks sparse/empty in the chat widget, so this is the default for any "走势/行情/表现如何" style question — including casual phrasing like "最近一周走势" (interpret as "show me how it's been doing lately", not literally 5 trading days).
   • Only request a shorter window when the user gives an unambiguous analytical ask for an exact short range (e.g. "对比昨天和前天的收盘价" or "最近3个交易日的具体数字"). Map common asks: 1个月→days:20, 3个月→days:60（默认）, 6个月→days:120, 1年→days:250.
3. For time-series results (price history, macro trend), must follow with show_widget:
   • Prefer structured JSON (Desktop renders via ECharts):
   • The top-level "type": "stock_chart" field is MANDATORY. Omitting it makes the Desktop client treat the payload as a generic HTML widget and render the raw JSON text instead of a chart — always double-check this field is present before calling show_widget.

Single stock:

jsonCopy code
{
  "type": "stock_chart",
  "title": "火炬电子 603678",
  "chart_type": "candlestick",
  "data_source_label": "获取数据 | AkShare（免费行情）",
  "points": [
    {"date": "2026-07-01", "open": 80, "high": 90, "low": 78, "close": 85, "volume": 120000}
  ],
  "attribution": "数据来源：AkShare"
}

Multiple focused stocks (Kimi-style tabs — user can switch 火炬电子 / 祥鑫科技 / 天齐锂业):

jsonCopy code
{
  "type": "stock_chart",
  "data_source_label": "获取数据 | AkShare（免费行情）",
  "attribution": "数据来源：AkShare",
  "watchlist": [
    {
      "symbol": "603678.SH",
      "name": "火炬电子",
      "chart_type": "candlestick",
      "points": [{"date": "2026-07-03", "open": 81.26, "high": 89.4, "low": 77.93, "close": 85.48, "volume": 42942215}]
    },
    {
      "symbol": "002965.SZ",
      "name": "祥鑫科技",
      "chart_type": "candlestick",
      "points": [...]
    },
    {
      "symbol": "002466.SZ",
      "name": "天齐锂业",
      "chart_type": "candlestick",
      "points": [...]
    }
  ]
}

When the user mentions several tickers they are tracking, fetch each via query_data_source and pass one show_widget with a watchlist array — do not render three separate widgets.

CRITICAL — never hand-roll HTML/ECharts for stock charts. Always pass the structured stock_chart JSON (single points or watchlist). The Desktop client has a dedicated, theme-aware StockChartWidget (dark/light adaptive tabs, 红涨绿跌, quote header). Hand-written <div>+ECharts <script> HTML produces broken results (e.g. white tab text on white background, sparse charts) and must not be used for stock/candlestick data.

Copy the OHLCV rows from the query_data_source result verbatim into points — the result is already trimmed to date/open/high/low/close/volume and the full window (e.g. 60 rows) fits, so there is no need to re-query with a smaller days. If a result ever looks truncated, use whatever rows you have; do NOT loop re-querying the same symbol.

• Macro trends: same shape with "chart_type": "line" and points like {"date": "2021", "value": 8.1}.

4. Workflow (same as show_widget discipline): 1–3 sentences visible intro → query_data_source → show_widget → interpret numbers from tool output only.

Example calls

Map query_data_source OHLCV rows to stock_chart.points (normalize date/open/high/low/close/volume field names from the JSON).

Fallback discipline

• MissingCredentialError or connection failure on ifind / tushare → try free sources (akshare, world_bank, imf) before giving up.
• Tushare without MCP → tell user to connect tushareMcp in Desktop settings; do not invent prices.
• If all applicable sources fail: say explicitly 「当前数据源暂不可用，无法核实最新数据」. Never substitute a remembered or guessed number.

Known plugins (authoritative list: list_data_sources)