We're happy to share several substantial improvements to the QuantConnect Object Store to make working with objects easier. These improvements improve the Object Store's user interface, speed, and redundancy.

The Object Store is a key-value data store that is accessible anywhere in an organization for machine learning models, debugging, or storing results and state for future analysis. 

Clients often use the object store to save trained ML models, which are picked up seamlessly from the live trading environment. Models can be serialized and reloaded from Jupyter research notebooks into live trading strategies. Other clients use the object store to restore their algorithm state between live-trading deployments by saving key indicators or signals. 

Object Store Video

The improvements include:

New User Interface: We have deployed an interactive UX where you can upload, browse, preview, and delete objects from your object store. Previously, working with the Object Store required custom notebooks, and users had little visibility into the objects they were creating. Now, you can quickly navigate and delete objects in bulk. This interface is located under the "Organization\Object Store" tab.

Data Distribution: Data distribution speeds and resilience have been improved to operate near theoretical limits. Models produced in research and backtesting environments will be accessible within seconds.

Improved Redundancy: We have built and tested new redundancy systems for the Object Store to strengthen the live trading environment. These new systems fall over within seconds in live trading, ensuring your live strategy stability.

We've also rolled out the feature to the QuantConnect API and the LEAN CLI to help you automate uploads of custom data types or locally trained models.

We're excited to continue improving the technology and will soon support backtesting using data sourced from your organization's object store. Using the object store for custom data will likely result in 10x speed improvements for users harnessing custom data. 

Happy Coding!

QuantConnect Team