Data Fetcher

Data Fetcher Skill

Purpose

Pull historical (and near-realtime) OHLC market data from external sources and load it directly onto the Vibe Trade chart canvas. Once loaded, every other skill — Pattern detection, Strategy generation, backtesting — can operate on the new dataset without the user having to upload a CSV.

This skill is the bridge between "I want to analyze X" and the rest of the platform's tooling.

When to use this skill

Vibe Trade should dispatch to the Data Fetcher when the user wants to:

• Load a stock/ETF/index ("Fetch AAPL daily", "Get SPY weekly data", "Load TSLA hourly for the last month")
• Load a crypto pair ("Fetch BTC/USDT 1h", "Get ETH 4h for last 1000 bars", "Pull DOGE 5m last 24 hours")
• Refresh recent data ("Get the last 100 bars of BTC", "Latest hourly data for Apple")
• Compare assets ("Load BTC and ETH" — fetch both as separate datasets)

Data sources

The skill auto-detects the right provider from the symbol shape:

• Pair separators (/ or - with crypto quote) → ccxt
• ^INDEX or =X → yfinance
• Bare crypto bases (BTC, ETH, ...) → ccxt with USDT quote
• Everything else → yfinance

Inputs

Outputs

Returns a SkillResponse with:

• reply — short confirmation ("Loaded 1000 bars of BTC/USDT 1h from binance.")
• data.dataset — the full fetched payload (symbol, source, interval, bars, metadata)
• tool_calls:
  1. data.dataset.add with the fetched payload — registers the dataset in the store and switches the chart to it
  2. notify.toast with a success message
  3. bottom_panel.activate_tab if a downstream skill has matching tabs

Tools used

Examples

Crypto pair

"Fetch BTC/USDT hourly data, last 500 bars."

→ Calls data.fetch_market(symbol="BTC/USDT", interval="1h", limit=500), auto-routes to ccxt:binance, returns 500 hourly bars, emits data.dataset.add to load them onto the chart.

Stock

"Get AAPL daily for the last 2 years."

→ Calls data.fetch_market(symbol="AAPL", interval="1d", limit=504) (504 ≈ 2 trading years), auto-routes to yfinance, loads onto the chart.

Bare crypto base

"Pull ETH 4h."

→ Auto-completes to ETH/USDT, fetches via ccxt:binance at 4h timeframe, default 1000 bars.

Underlying implementation

Wired through core/agents/processors.py::_data_fetcher_processor, which:

1. Parses the user message via core.data.fetcher.parse_query() to extract symbol, interval, and limit
2. Calls core.data.fetcher.fetch() to pull bars from the right provider
3. Wraps the result in a SkillResponse with the right tool_calls

The fetcher in turn uses:

• yfinance for stocks/ETFs/indices/forex
• ccxt for crypto, defaulting to the Binance public API

Both libraries are pip-installable, key-less, and ship with the backend.