Perplexity’s Portable Computer agent now runs on Windows PCs equipped with Nvidia RTX GPUs featuring 24GB or more VRAM. This marks a key step in delivering autonomous agent workloads locally while leveraging hybrid cloud compute for heavier tasks.
- Runs locally on Nvidia RTX GPUs with ≥24GB VRAM on Windows
- Hybrid architecture escalates complex tasks to cloud with user consent
- Supports enterprise integrations like Outlook, Slack, GitHub via connectors
Infrastructure signal
Perplexity’s Windows agent demands high-end Nvidia GPUs with at least 24GB VRAM, targeting RTX and RTX PRO series. This hardware dependency highlights the resource intensity of running a 27-billion-parameter LLM locally with autonomous agent capabilities. The model runtime includes specialized integrations for GPU orchestration and sandbox security to isolate workloads.
Developer impact
Developers gain a more autonomous local agent environment capable of accessing both local files and external APIs such as Microsoft Outlook, Google Drive, Slack, and GitHub. This increases workflow efficiency by enabling multi-step tasks and tool calling within the same machine, reducing the need to offload everything to the cloud and addressing data residency concerns.
However, developers must navigate increased complexity in runtime orchestration and security sandboxes that manage hybrid task execution and resource access. The scaling from experimental setups to enterprise-grade tools requires robust platform engineering to maintain consistency across GPUs, OS variations, and cloud integration points.
What teams should watch
Teams deploying or developing agent-centric solutions on cloud native infrastructure should closely track Perplexity’s hybrid local/cloud model approach. Ensuring GPU availability with specific VRAM thresholds and understanding the cost impacts of cloud escalations versus local compute will be essential for budgeting and architecture decisions.
Security and compliance teams need to vet how the agent manages sensitive data locally in a sandbox versus transmitting to cloud models only with explicit user consent, as this hybrid data flow paradigm may influence enterprise policies around data sovereignty and regulatory controls.