Databento API guide

Schemas and datasets

Pick a dataset (one venue feed) and a schema (level of detail). Historical requests are priced by data volume, and metadata.get_cost tells you the cost before you download. Databento reports roughly 6 µs normalization latency for live data, a vendor processing figure rather than end-to-end latency.

import databento as db
client = db.Historical("db-YOUR_KEY")
data = client.timeseries.get_range(dataset="GLBX.MDP3", schema="ohlcv-1d",
    symbols="ES.c.0", stype_in="continuous", start="2024-01-01", end="2024-12-31")
print(data.to_df().tail())

Build a Databento strategy with AlgoGP

AlgoGP does not connect to Databento natively yet. Databento uses plain HTTPS with an API key, so it is a good candidate for a native adapter; it is not built in yet.

1. 1

   Design and backtest in AlgoGP

   Describe the strategy to the copilot and backtest it on years of normalized history. The result does not depend on any vendor.

2. 2

   Export the Python project

   Every project exports strategy.py, market_data.py, brokers.py and live.py. The strategy only sees bars, cash, positions and orders, never a vendor’s wire format.

3. 3

   Plug in Databento

   Implement the four adapter methods (account, positions, submit_market_order and a candle fetcher) with Databento’s SDK, or keep a built-in platform for paper trading and add Databento when you go live.

Built-in data sources for the same markets: Alpaca, Yahoo Finance, Alpha Vantage, Tradier.