Perplexity launches Hybrid Compute for Mac, splitting tasks between cloud and local AI

Hybrid Compute is available to Pro, Max and Enterprise subscribers on any Apple silicon Mac running macOS 15 or later with at least 24GB of unified memory (32GB recommended). The local model installs in one click from the Mac app, with no Ollama, no separate runtime and no API key, and local processing consumes no cloud credits.

Perplexity said every task begins in the cloud, where frontier models handle web search, planning and long-horizon reasoning. When a step touches private files or sensitive data, the system hands that step down to the local model without restarting the task or losing context, then merges both halves into one result. This reverses the local-first compute mode Perplexity shipped on NVIDIA DGX Spark a week earlier, which starts on user hardware and escalates to cloud models with permission. Because Computer works with iPhone, a task can be triggered remotely while sensitive steps execute on the Mac.

Before anything from a protected file reaches the cloud, an on-device classifier inspects it and the gate applies one of four outcomes: keep it local, mask the sensitive spans, refuse the action, or ask the user for consent. Credentials, payment card numbers and government IDs get the strictest handling. Masked values are swapped for stand-ins on the way out and restored when the cloud answer returns.

Perplexity also open-sourced the classifier behind that gate. PII-Tracer is a 0.6B-parameter bidirectional encoder adapted from a Qwen3 backbone, with a linear tagging head that emits 37 labels for nine PII types and an auxiliary head that predicts whether a conversation contains sensitive material. The model was trained for three epochs on roughly 714,000 samples. The accompanying benchmark, PII-TRACE, contains 13,148 synthetic conversations across 13 languages and 10 writing systems, with 37,431 character-level identifier mentions.

Perplexity said that among 12 detectors tested, PII-Tracer achieved the highest character F1 (0.629) and the second-best span-overlap and span-containment F1, behind GPT-5.6-sol. It led on consistency, finding every mention for 79.4% of recurring identifiers and 77.6% of cross-turn identifiers, versus 57.0% and 55.1% for GPT-5.6-sol. Single-window recall drops from 0.975 on conversations under 1,000 characters to 0.687 at 10,000 characters or more; sliding-window decoding with 50% overlap lifts overall character recall from 0.830 to 0.965 on the same checkpoint.

Hybrid Compute supports three local models at launch: Gemma 4 E4B, Qwen3.6 35B-A3B, and a Perplexity model post-trained for Computer. On the cloud, users can choose more powerful models such as Claude Opus 5 or GPT-5.6 Sol, according to Android Authority. The same outlet reported that users will be notified if personal data appears in files or attachments for a task, and can then decide whether to split the task, keeping private files local while running the rest in the cloud, or upload everything anyway. Because local processing does not incur token costs, the hybrid approach can also reduce overall usage costs.

For Enterprise customers, administrators can set org-wide rules for what must stay on device, what may be masked and what requires explicit approval, and they receive audit logs for when information leaves a machine. Perplexity positions the feature as usable for legal, healthcare and financial teams, where the context that makes agentic assistants useful is often too sensitive to send to a cloud endpoint.