Home AIPerplexity’s Hybrid Compute Keeps Sensitive Work on Local LLMs

Perplexity’s Hybrid Compute Keeps Sensitive Work on Local LLMs

Perplexity Hybrid Compute: cloud brains for the hard parts, local LLMs for the secrets

by Roronoa Zoro
0 views

Perplexity is shipping Hybrid Compute for Perplexity Computer: a workflow that can split a job between a frontier cloud model (examples cited include Opus-class and GPT-5.6 Sol) and a local LLM so confidential material never has to leave the machine. Engadget reports a new privacy classifier that flags files and prompts you should keep local, with local model options including Gemma E4B and Qwen 3.6 35B variants — more local choices promised later.

Hybrid Compute is Mac-first for now (Apple Silicon, macOS 15, 32GB unified memory recommended). Product lead Jon Staff’s framing is user agency: the system suggests a split, but people still decide what crosses into the cloud. For teams that want agentic research tools without a blanket “upload everything” policy, this is the practical middle path.

Source: https://www.engadget.com/2248548/perplexitys-hybrid-compute-splits-sensitive-tasks-between-cloud-and-local-ai/