If you haven’t checked it out yet, here is our July newsletter! 👇
A lot happened on Tensorlake sandboxes last month, and we have even more coming in August. Stay tuned!
Tensorlake
@tensorlake
Scalable Sandbox and Orchestration Infrastructure for Agents at
tensorlake.ai
- Announcing Tensorlake BYOC - serverless sandboxes in your private networks on AWS, GCP, Azure and other neo clouds. This enables running sandboxes which can connect to your internal infrastructure and systems. You can run upto 5 million concurrent sandboxes per Tensorlake
- Devin from @cognition can now run on Tensorlake's SSD backed Sandboxes via Outposts. Our Outpost support can suspend and resume sandboxes when a conversation thread is idle without losing its file system or memory state. We bring the sandbox back to life whenever Devin needs to
- Thanks, @YossiEliaz and @steipete, for adding Tensorlake sandbox support to Crabbox! crabbox.sh crabbox.sh/providers/tens…
- Really cool to see @tensorlake used as an execution backend for this research. The paper explores an important challenge with parallel AI agents: running more agent branches creates more executions, but not necessarily more independent evidence. Different agents may still relyMicroVM sandboxes are the commodity compute unit of the AI era. The Boltzmann MapReduce paper shows how to reduce over forked ensembles using partition functions (precision-weighted pooling) instead of naive averaging. Ran on Islo, Daytona (1,024 forks), Tensorlake, and Modal,
