Market Watcher
Market Watcher answers how a stock, an index, a currency pair or a crypto asset has been doing — with a chart you can explore and a sentence or two saying what it shows.
At a glance
| Handle | @market-watcher |
|---|---|
| Availability | Seeded in every organisation |
| Tags | markets, charts, finance |
| Model tier | default |
| Autonomy | supervised (the platform default — the definition declares none) |
| Tools | market_data, render_chart, web_search, web_fetch, calculate, current_time |
| Max turns | 20 |
| Schedulable | Yes — a daily market summary is an automation |
How it works
-
Data first, by reference
market_datapulls end-of-day OHLCV history from keyless providers and registers it as a run dataset per symbol, with dimensionsdate, open, high, low, close, volume, sma20, sma50, pct. The model never retypes a price table. -
One chart, referencing that dataset
render_charttakes a full ECharts option that points at the dataset by name. A five-year candlestick with a volume pane costs a few hundred tokens instead of thousands of rows. -
The news behind a move
When the question is "why", it searches and reads, and cites what it used.
Which panes a price chart carries — candlesticks with a moving average, a volume pane beneath, a side-by-side performance comparison — lives in the agent's definition, not in code, for the same reason ITHelp's intake fields do: it is product judgement an organisation may want to change without a deploy.
How the router finds it
A request naming a tradeable instrument and a price or chart intent —
"chart 50 days of Tesla stock", "bitcoin price trend" — reaches whichever
visible agent carries the markets tag before any model is asked.
Both halves are required: "chart the weather next week" belongs to the
Weather Agent, which also draws charts,
and "what is our stock policy" belongs to
Procurement.