AI;DR
AI;DR
← Back to live feed1 story

Tuesday, Aug 25, 2026

1

Perplexity Ships First Local AI Agent Runtime for Nvidia DGX Spark to Hit 73% Bench Score

Portable Computer is a new local AI agent stack that executes the entire runtime—including the orchestrator LLM, subagent LLM, and agent harness—on-device using Nvidia DGX Spark. Available to Perplexity Pro and Max subscribers, the system initially supports the post-trained PPLX 27B model and Qwen 3.8 27B. The platform offers a one-click local inference setup to eliminate cloud dependencies for standard agentic workflows.

The system utilizes a hybrid inference model that routes complex reasoning tasks to cloud frontier models after manual user approval and PII filtering. This escalation increases the system's Terminal Bench 2.1 score from 59.6% to 73.0% at a cost of $0.415 per rollout. Support for Nvidia Nemotron 3.5 Lightning is planned, and remote models are restricted to returning text guidance without direct access to local files.

Sign in to suggest edits

Key sources

  1. SOURCE@perplexity_ai“entire runtime: orchestrator LLM, subagent LLM, agent harness all run on your local hardware”x.com
  2. SUPPORT@perplexity_ai“Your sensitive documents stay on your device”x.com
  3. SUPPORT@perplexity_ai“support for NVIDIA Nemotron 3.5 Lightning coming soon”x.com
  4. SUPPORT@perplexity_ai“available today for all Perplexity Pro and Max subscribers with an @NVIDIA DGX Spark”x.com
  5. SOURCE@nvidia“offers one-click local inference setup and an optimized agentic experience for DGX Spark”x.com
  6. SUPPORT@aravsrinivas“In a compute and power-constrained world, a good chunk of agentic inference needs to move to local hardware”x.com
  7. SUPPORT@aravsrinivas“escalation lifts the score from 59.6% to 73.0% at $0.415 per rollout”x.com
Markdown