Talos’s Coin Metrics Data now accessible via Amazon S3 and Snowflake Cloud Delivery

We're introducing Cloud Delivery, a new way to access Talos’s Coin Metrics Data. Instead of requesting data via the API, clients can now read whole datasets from their own environment, with 10 datasets published daily as Apache Parquet files in Amazon S3 and as queryable tables in Snowflake. 

Our users’ data consumption follows two predominant use cases. The first calls for a specific market over a specific interval, returned on demand. The API serves this use case through targeted queries, real-time streaming and access to the latest observation within milliseconds. The second calls for an entire dataset to be resident in the user's own environment, where it can be scanned repeatedly across successive pieces of analysis. For example, multi-year backtests, full order book reconstruction and data lake ingestion are bulk transfers that fall into this category which Cloud Delivery is well suited to address.

Availability in Amazon S3

Cloud Delivery publishes 10 datasets as Apache Parquet objects, refreshed daily. Users read these objects from their own AWS account using existing tooling, including the AWS CLI, the AWS SDKs, DuckDB, Spark, Athena and Glue.

Objects are organized using Hive-style key=value prefixes by market type, exchange and date:

s3://talos-data-daily/
  data-type=market-trades/
    market-type=spot/
      exchange=coinbase/
        dt=2026-07-28/
          *.parquet

Query engines discover these partitions automatically and eliminate prefixes that cannot satisfy a given filter, so a read scoped to a single exchange and date resolves against a small number of objects rather than the full dataset. Retrieving a day of trades requires no client library and no pagination:

aws s3 cp \
  s3://talos-data-daily/data-type=market-trades/market-type=spot/exchange=coinbase/dt=2026-07-28/ \
  ./trades/ --recursive --request-payer requester

Other query engine tools can equally read the prefix in place, without a download step.

Availability in Snowflake

The same 10 datasets are exposed as tables within the user's own Snowflake account through Secure Data Sharing. Users mount a read-only database, grant access to their reader roles, and query it using standard SQL. Partition columns are exposed as ordinary columns, so filtering is conventional:

SELECT dt, market, price_close, volume
FROM   coin_metrics.EXTERNAL.market_candles
WHERE  market_type = 'spot'
  AND  exchange    = 'binance'
  AND  frequency   = '1d'
  AND  dt BETWEEN '2026-01-01' AND '2026-06-30';

The database name is chosen by the user when the share is mounted. No pipeline development is required and no data is copied.

Dataset coverage

Ten datasets are published: market trades, candles, quotes, order book snapshots, order book updates, reference rates, and current catalogs for markets, assets, exchanges, and pairs. Coverage is as follows.

• Trades and candles. 48 exchanges, across spot, futures, and options. Candles at eight frequencies, from one minute to one day.
  
  

• Quotes and order book snapshots. 29 exchanges, across spot and futures, at raw granularity and full book depth.
  
  
• Order book updates. 5 exchanges, across spot and futures. Additional exchanges can be added on request.
  
  
• Reference rates. All covered assets, at one-second and one-hour frequencies, against USD, EUR, ETH and BTC.
  
  
• Reference data. All covered entities.

Each dataset is produced once daily and covers the preceding UTC day. Column names and definitions correspond to the fields returned by the API. For the market-level universe behind each dataset, see the Coverage Tool.

Access and onboarding

Cloud Delivery is available as a limited release, with general availability to follow. It is provisioned per user. Existing clients can contact their Talos sales representative or account manager. Teams not yet working with Talos can reach us at sales@talos.com.

Full documentation is available for Cloud Delivery, including per-destination guides for Amazon S3 and Snowflake, covering schemas, partition layouts, update schedules and query examples.

‍

‍