Perplexity adds a local AI mode for Mac

Perplexity launched Hybrid Compute for its Mac app on September 1, splitting each Perplexity Computer task between a frontier model in the cloud and a model running locally on the machine. A task starts in the cloud; any step that touches private files or sensitive data drops to local inference behind what Perplexity calls a privacy gate, and nothing from that step leaves the device.

The default local model is PPLX Qwen 3.8 27B, a 27-billion-parameter model Perplexity post-trained from Qwen 3.8 specifically for Computer workloads, installed in one click. Two more local options ship alongside it: Gemma 4 E4B and Qwen3.6 35B-A3B. Work done locally draws no cloud credits and needs no API key, which makes it free to run on hardware you already own once the model is downloaded.

Hybrid Compute requires Apple silicon on macOS 15 or later, with 24GB of unified memory as the floor and 32GB recommended; Perplexity singles out the new Mac mini and Mac Studio as the platforms it is tuned for. It’s available now to Pro, Max, and Enterprise subscribers. The move sits in the same on-device push as Prism ML’s Bonsai 27B running on iPhone and Nvidia’s portable-computer partnership with Perplexity: frontier labs are increasingly willing to trade some capability for keeping an operator’s private data off their own servers entirely, not just encrypted in transit. Anyone routing agent workloads by cost and privacy rather than raw benchmark score should treat this as a third lever alongside model choice, next to this week’s Gemini 3.8 Flash release.