Perplexity launches SPACE, a secure sandbox for its AI agents
The new platform runs agent tasks inside isolated micro-VMs, letting them pause and resume work across days without exposing your credentials.
Perplexity has announced SPACE, a sandbox built to let its AI agents work at full capability while keeping the process contained. According to SiliconANGLE, the platform launched on 15 July 2026.
The headline ideas are security and persistence. Each task runs in its own isolated micro-virtual-machine (using AWS’s Firecracker technology), and any passwords or keys an agent needs stay isolated per session and under the user’s control. Agents can also pause, resume and branch long-running work across sessions that last hours, days or even a week, picking up where they left off rather than starting over.
Why it matters: the biggest barrier to trusting an AI agent with real work isn’t intelligence, it’s safety — you don’t want it loose in your accounts. Isolating each task and ring-fencing credentials is exactly the kind of plumbing that has to exist before “agents that do things for you” become something sensible people rely on. It’s worth watching whether rivals follow with similar guardrails.
Source: original announcement ›