Awesome Infra for AI › Workflow Orchestration for AI

tensorlakeai/tensorlake

⭐ 1016 Python added to this list on 2026-08-17 repository created 2024-11-15

Tensorlake is a compute infrastructure platform for running agentic applications. Its Sandbox API creates stateful Firecracker MicroVMs that serve as isolated execution environments — a place to run an agent itself, or to execute tools and model-generated code without exposing the host. The sandboxes are built for speed and statefulness: block-based storage gives near-SSD filesystem throughput inside the VM, creation takes under a second through a dynamic cluster scheduler, and snapshots capture memory and filesystem at any point so an environment can be checkpointed, cloned instantaneously across machines, or suspended when idle and resumed in under a second with no loss of state. Live migration moves running sandboxes between hosts during maintenance with only a brief pause, and a single project can scale to millions of sandboxes. Network policy is part of the sandbox surface: outbound access can be blocked entirely, restricted to an allowlist of destinations, or restored, on a running sandbox. Alongside the VMs, Tensorlake provides a serverless function runtime with fan-out, which supplies the long-running orchestration layer agents need when a task outlives a single request — durable execution, parallel branches and result aggregation. Access is through a Python SDK installed from PyPI and a standalone tl command-line binary that creates sandboxes, executes commands inside them, copies files in and out, opens an interactive shell and adjusts network policy. The platform is a hosted cloud service with an API key, and the repository holds the SDK and CLI. It targets teams building agents that need real isolation and durable state rather than a process on a shared host.

https://github.com/tensorlakeai/tensorlake

sandboxmicrovmfirecrackeragent-runtimeorchestrationserverlessisolation