Perplexity: AI agent can now run fully locally on a Mac

Perplexity says in an announcement dated October 20, 2026, that its AI agent Perplexity Computer can now run fully locally on a Mac. Both the language model itself and the agent that carries out the tasks run on the machine, and according to Perplexity, no data is sent to cloud servers. For now, this is only Perplexity's own statement, and no independent coverage of this particular news has been found.

In practice, this means sensitive work can stay on the Mac throughout the task, according to Perplexity. The agent can work across the Notes, Messages, Mail and Numbers apps. Perplexity cites tasks such as portfolio analysis, drafting proposals and prioritizing tasks.

Requirements and price

The local mode uses either Qwen 3.6 or PPLX Pro, a version of Qwen that Perplexity has post-trained itself. It requires an Apple silicon Mac and the latest Perplexity Mac app. According to the source text, M5 Pro, M6 or newer is specified, and 36 GB of RAM or more is recommended. However, this should be checked against Perplexity's own site before buying anything.

The mode is available from October 20, 2026, for Pro, Max and Enterprise subscribers. No separate price has been mentioned for it.

The cloud is still an option

Running locally has limits in computing power compared with the cloud. If a task requires current information, browser use or connected apps, the user can give permission for it to go to the cloud.

Hybrid mode from September

Perplexity already launched Hybrid Compute in the Mac app on September 1, 2026. It splits a task between advanced models in the cloud and a local model on the Mac. The cloud handles reasoning, web search and planning, while the local model handles private files and actions on the device.

An on-device so-called privacy gate is meant to recognize sensitive information such as names, addresses and account numbers before it leaves the Mac. It can mask the information, keep it local, reject the action or ask for consent. According to Perplexity, Hybrid Compute requires macOS 15 or newer and at least 24 GB of unified memory. Engadget cites 32 GB as the recommended minimum.

According to Engadget, there is no charge for tokens generated by the local models. Perplexity itself admits that a clean result from the large cloud models will almost always be better at raw artifact creation. The company points to lower costs and better privacy as a counterweight.