Perplexity has launched Portable Computer, a fully local version of its AI agent platform, running on user hardware with no cloud dependency. The service is initially available to Pro, Max, Enterprise Pro, and Enterprise Max subscribers.
TL;DR
- Perplexity introduces Portable Computer, a local version of its AI agent platform, running on Nvidia hardware.
- The service is initially available to paid subscribers and offers zero per-token inference costs and enhanced privacy.
- Portable Computer supports Qwen 3.8 27B and PPLX 27B models, with Nvidia's Nemotron 3.5 Lightning coming soon.
What happened
Perplexity has launched Portable Computer, a fully local version of its cloud-based AI agent platform Computer. The new product runs the orchestrator LLM, subagent LLM, and agent harness entirely on user hardware, initially supporting Nvidia's DGX Spark workstation and Linux systems with RTX GPUs. Windows support is expected in September, requiring at least 24GB of VRAM.
Available models include Qwen 3.8 27B and PPLX 27B, with Nvidia's Nemotron 3.5 Lightning coming soon. The service is limited to Pro, Max, Enterprise Pro, and Enterprise Max subscribers. Perplexity highlights zero per-token inference costs and enhanced privacy since data stays on-device unless users grant cloud access.
Why it matters
Portable Computer addresses growing concerns around inference costs, data privacy, and the limitations of cloud-dependent agentic systems. Local inference avoids per-token API fees, which can accumulate quickly for users running complex agentic workflows in the cloud.
The launch comes amid reports of Nvidia's potential $30 billion investment in Perplexity, positioning Portable Computer as a significant strategic move. The high hardware requirements and limited model selection may impact adoption, but Perplexity is betting on the value of local control, privacy, and predictable costs.
Key facts
- Portable Computer runs entirely on local hardware, with no cloud dependency by default.
- Initial hardware support includes Nvidia's DGX Spark workstation and Linux systems with RTX GPUs.
- Windows support is expected in September, requiring at least 24GB of VRAM.
- Available models include Qwen 3.8 27B and PPLX 27B, with Nvidia's Nemotron 3.5 Lightning coming soon.
- The service is limited to Pro, Max, Enterprise Pro, and Enterprise Max subscribers.
- Perplexity highlights zero per-token inference costs and enhanced privacy with data staying on-device.
- Benchmark results show Portable Computer matching or exceeding the accuracy of competing agent harnesses.
- Portable Computer can connect to external services like Slack, Google Drive, Gmail, and GitHub with user permission.
Context
The launch of Portable Computer reflects a broader industry trend toward edge deployment of AI models. Nvidia has been aggressively pushing its Jetson line for robotics and physical AI applications, emphasizing optimized, compressed large models for real physical devices.
For Perplexity, which has experimented with search, advertising, browsers, and personal assistants, Portable Computer represents another attempt to carve out a distinct position in an increasingly crowded AI landscape. The move underscores the company's strategic focus on addressing key user concerns around cost, privacy, and control.