Perplexity demonstrates fully local agent runtime on DGX Spark

AravSrinivas · x · 2026-08-25

Perplexity showcased a fully local agent runtime enabled by open-weight models. They adapted the cloud-based Perplexity Computer harness to a local-first version, orchestrated by a local 27B MoE running on NVIDIA DGX Spark. This moves agentic inference to local hardware in a compute-constrained world.

Related event: Perplexity Teams Up with NVIDIA on Fully Local Agent Stack 'Portable Computer'(22 posts)→

Original post →

More from coding & agent

coding & agent channel →