Overview
- Perplexity rolled out a fully local mode for its Mac app Computer on Tuesday, October 20, 2026, that can run both the AI model and the controlling agent on the Mac so tasks can be completed offline.
- Users enable the feature by downloading a local model in the Perplexity Mac app and choosing the fully local option, and the on‑device model choices at launch are Qwen 3.6 and Perplexity’s PPLX Pro.
- Perplexity recommends roughly 36 GB of unified memory for the best local performance, a guideline that restricts practical use to higher‑end or newer Apple silicon machines and leaves less capable Macs to rely on Hybrid Compute.
- With user permission the local agent can access native apps and files and runs code and tools inside isolated sandboxes, but Perplexity has not clarified whether macOS Full Disk Access or other specific permissions are required.
- The change lets users keep private data on their Macs for lower latency and reduced cloud exposure while still allowing them to call cloud frontier models for up‑to‑date information or harder reasoning when they choose, though many users may lack the required hardware to run the mode locally.