Use Databento data for backtesting with your AI coding agent.

How does Databento access work for backtesting data?

You sign up on the Databento website and API keys are issued automatically. Official client libraries exist for Python, Rust and C++. The client reads the key from an environment variable, so the key never sits in your code.

Live data needs a paid subscription (reported). Plans, prices and licences are set per dataset (reported). A CME futures plan does not automatically cover equities or options, so check coverage for your venue.

Backtesting and paper runs with Databento data

Historical data comes as time-range requests or batch file downloads, in the same schema as the live feed. Instrument definitions are point in time, which helps avoid look-ahead bias. Futures roll symbols map to instrument ids that change at each roll, so store the mapping with its date.

Databento is data only, so it has no paper trading and places no orders. New users get USD 125 of historical credits through the same API as a paid account (reported). The credits expire after 6 months, and there is no free live tier.

For a paper run, Super Algo Kit pairs the feed with a broker card such as Alpaca or Interactive Brokers. Any paper or backtest run on historical replay is labelled as simulated, with the data source named.

Databento limits and costs to plan for

Live sessions are capped per dataset (reported). The cap is 10 on Standard and 50 on Plus and higher, in force since 16 February 2025. A restart loop that leaks sessions can lock you out. Databento reportedly calls historical limits practically unlimited for a well-behaved app; the numbers are not confirmed.

Historical data is billed per GB of uncompressed size (reported). An order-by-order request over many symbols can use the free credits fast. Standard live data is reported at USD 199 a month; Plus and Unlimited add larger licence fees on annual contracts. These prices came from page summaries, so check the pricing page first.

What Super Algo Kit does with Databento

The adapter plans one live session per dataset per process and at most 5 sessions in total on Standard. It backs off on any refusal. These caps are the kit's own settings, below the vendor's limits.

Event time and receive time are kept apart, and a record without an event time is labelled receive-only. Order books are rebuilt from order-by-order data with Databento's own library, not by hand. Prices are never compared as floating-point numbers until their scaling is confirmed.

Feed health is three separate checks: connection state, tick age and coverage of your instruments. A closed market is not treated as a stale feed. After the close, the bars the kit built are checked against Databento's own bars.