Overview
- The system breaks user goals into subtasks, assigns them to sub-agents, and can operate asynchronously for hours, days, or months with persistent memory.
- Tasks run in isolated compute environments with a real browser, filesystem, and curated tool integrations, positioning it as a safer alternative to local agents such as OpenClaw.
- Perplexity currently routes work across about 19 models, using Claude Opus 4.6 for core reasoning alongside tools like Grok, Google’s Nano Banana and Veo 3.1, and GPT‑5.2 for long‑context recall and wide search.
- Computer is available now to Perplexity Max subscribers under usage-based credits with optional spending caps, with staged expansion to Pro and Enterprise users planned.
- The debut signals a pivot from AI search to an execution platform focused on delivering finished outputs through multi-model orchestration.