Flat-File Access with AWS CLI

1. Export The Credentials

export AWS_ACCESS_KEY_ID="<TEMP_ACCESS_KEY_ID>"
export AWS_SECRET_ACCESS_KEY="<TEMP_SECRET_ACCESS_KEY>"
export AWS_SESSION_TOKEN="<TEMP_SESSION_TOKEN>"
export CHD_BUCKET="cryptohftdata-processed"
export CHD_ENDPOINT="https://e5de051a8af1332b58a11bd5b2e40258.r2.cloudflarestorage.com"

2. List The Bucket

This confirms that the temporary credentials, endpoint, and session token are all accepted by the R2 S3-compatible API.

aws s3 ls "s3://$CHD_BUCKET/binance_spot/" \
  --endpoint-url "$CHD_ENDPOINT" \
  --region auto

3. Download A Known Object

aws s3 cp "s3://$CHD_BUCKET/binance_spot/2025-08-01/20/BTCUSDT_trades.parquet" ./BTCUSDT_trades.parquet \
  --endpoint-url "$CHD_ENDPOINT" \
  --region auto

4. Download All Data For One Date

Object keys follow the pattern {exchange}/{YYYY-MM-DD}/{HH}/{symbol}_{data_type}.parquet. To pull every hour for one exchange on one date, copy that date prefix recursively.

Parquet files carry ZSTD compression inside the file, so they open directly in pandas, Polars, or DuckDB with no separate decompression step. Objects collected before 19 August 2026 are stored under a .parquet.zst key instead and need zstd -d first. S3 access reads the bucket directly, so the key has to match exactly; the /download API accepts either spelling for any date.

aws s3 cp "s3://$CHD_BUCKET/binance_spot/2025-08-01/" ./binance_spot_2025-08-01/ \
  --recursive \
  --endpoint-url "$CHD_ENDPOINT" \
  --region auto