From whichever vendor you choose, in one call. Reamer Research's bundled fetchers pull historical bars from your own Alpaca or Databento account straight into a backtest, and bars of any type from any other source go in as arrays or as memory-mapped files built from CSV. You pick the vendor, pay it directly and switch whenever you like, and the engine stays lean because it carries no data feed of its own. Reamer Server takes orders from your strategies, which price from your own feed.
The kit includes synthetic sample bars inside Evaluation Mode, so the quickstart runs on its first try.
Getting bars into Reamer Research
- From Alpaca or Databento.
python/fetchers/downloads historical bars from your own Alpaca account (US stocks and crypto) or Databento account (OHLCV, including continuous futures) straight into a backtest, andexamples/fetch_and_run.pydoes it in one command. Alpaca crypto needs no account. Alpaca stock bars arrive adjusted for splits and dividends. Each Databento request is priced first, and one over your limit is refused. - Arrays from any source. Load bars from CSV with pandas or numpy, a database or a vendor API, and pass them in as arrays: timestamp, open, high, low, close and volume for each bar. This works from Python, C++ or any language that can call C.
.binfiles.bin/reamer-csv-build AAPL.csv AAPL.binconverts a CSV. It detects the delimiter, columns and timestamp format, and reports what it skipped. Timestamps are whole seconds, so two rows at the same timestamp stop it unless--keep-firstkeeps the first bar of each second. It needs no license. Pass.binpaths from Python or C++, or callreamer_run_backtest_files(), and the files are memory-mapped, so a dataset larger than RAM runs, and a start and end time select a date range without rewriting them.- One instrument per file. If a CSV holds several symbols, the converter stops until
--symbolselects one. - Any bar type. Time, tick, volume, dollar, imbalance or your own custom rule. You build the bars, and the engine treats every type the same way. OHLCV is the record format, not a limit on how a bar was built.
- Several instruments at once. Each instrument has its own series, and timestamps do not need to line up across them. Every instrument is cut to the range they all share, so the shortest series sets the run's length.
- Non-price data, such as earnings dates or an index level, can be attached to each instrument from Python, C++ or the C interface. A bundled tool converts it from a two-column CSV.
What the engine expects from your data
- Data quality. A bar with a NaN or infinite price stops the run with an error. Wrong prices, missing bars or bad prints do not; they produce wrong results.
- Corporate actions. Unadjusted equity data gives wrong results with no warning.
- Futures rolls. Supply a continuous series. If it is unadjusted, you can declare the roll months (
roll_cycle_monthsin Python), and the engine stops treating the jump at each roll as a real price move.
See How do I backtest a strategy that uses earnings dates or other non-price data? and What is Reamer Research?