Perplexity Unveils Space, a Secure Sandbox Platform for AI Agents (techstrong.ai)

Perplexity has announced SPACE, a secure sandbox platform specifically designed for running AI agents. Unlike traditional sandboxes, SPACE utilizes Firecracker microVM technology to isolate tasks, ensuring that any potential failures or exploits remain contained. This allows long-running tool-using agents to perform complex, multi-step workflows without compromising system integrity or exposing sensitive credentials, significantly enhancing security in AI operations. The significance of SPACE lies in its ability to treat AI agents as first-class applications rather than simple chatbots. By employing a zero-trust model, each agent operates within its own minimal environment, thus minimizing cross-tenant risks and preventing unauthorized access to API keys or underlying systems. SPACE’s architecture includes a Control Plane for managing sandboxes, Node-level Services that manage credential access, and an efficient in-sandbox layer leveraging caching and Btrfs for rapid creation and restoration of sandboxes. This set-up not only optimizes performance but also reinforces security, positioning SPACE as a robust solution for deep research, data analysis, and autonomous report generation in the AI/ML community.