Today we’re launching Portable Computer on @NVIDIA DGX Spark.Portable Computer is a fully local version of Perplexity Computer, where the entire runtime: orchestrator LLM, subagent LLM, agent harness all run on your local hardware. No cloud dependency. pic.twitter.com/plVWz5PaAw— Perplexity (@perplexity_ai) August 25, 2026 A Full AI Agent Moves Onto the DevicePortable Computer is designed to bring the broader capabilities of Perplexity’s cloud-based Computer agent directly onto local hardware. Rather than offering only an offline chatbot, the system includes the models, inference engine, agent harness, tools, connectors and security sandbox required to complete multi-step tasks.Perplexity said the local agent can analyze files, search documents, work with code and execute actions on the device. The company has positioned this as a simpler alternative to assembling a local AI stack manually. According to Perplexity’s launch announcement, the orchestrator, planner, tool router, scheduler, task queue and local search index can all operate on the device.Local Models Handle the WorkloadPortable Computer’s local architecture is built around models that can perform much of the agent’s workload without sending data to remote servers. At launch, users can run Qwen 3.8 27B or PPLX 27B, a Perplexity post-trained version of the Qwen model. Nvidia’s Nemotron 3.5 Lightning is also planned for the model selection.Perplexity said the local model is trained to complete as much of a task as possible on-device, including reading files, searching documents and taking actions. The system is initially optimized for Nvidia’s DGX Spark, which combines a 20-core Arm CPU, Nvidia GPU and 128GB of unified memory.How It Manages the Workload on Local DevicesThe key to Portable Computer is that the entire agent workflow does not have to depend on a cloud API. The local orchestrator determines how a task should be handled, while the local model performs inference and accesses files and tools on the machine. Perplexity said the system can keep jobs running and use a local search index to retrieve information from files and code.For more demanding work, the agent can escalate individual steps to a cloud model, but it asks the user for permission before sending information outside the device. This creates a hybrid architecture in which routine processing remains local while frontier models are reserved for tasks that require additional reasoning or current web information.