Quick Answer
Perplexity Hybrid Compute keeps designated sensitive data on an eligible Mac while cloud models handle web search, planning, and frontier reasoning. Perplexity says its privacy gate can mask data, keep it local, refuse an action, or ask permission before cloud transfer. The feature requires a qualifying Apple-silicon Mac, so confirm your memory and macOS version before enabling it.
Key Takeaways
- Perplexity Hybrid Compute became available in the Perplexity Mac app on September 1, 2026.
- The feature is available to Perplexity Pro, Max, and Enterprise subscribers.
- Hybrid Compute uses cloud models for search and reasoning while a local model handles private files and on-device actions.
- The feature requires an Apple-silicon Mac running macOS 15 or later with at least 24GB of unified memory.
- Enterprise administrators can set rules for data that must remain local, can be masked, or needs user approval before cloud transfer.
What is Perplexity Hybrid Compute on Mac?
Perplexity Hybrid Compute is a Mac feature that divides an AI task between a local model on the computer and cloud-based models. Perplexity introduced the feature on September 1, 2026, in its Mac app for Perplexity Pro, Max, and Enterprise subscribers. Perplexity’s product announcement says the cloud portion handles web search, planning, and frontier reasoning, while the local portion handles private files, sensitive information, and on-device actions.
Perplexity Hybrid Compute matters because an AI request can involve both information that benefits from online research and information that a user may not want to upload. A task could require the assistant to search the web for current material while also reviewing a private document stored on the Mac. The feature is designed to keep those parts of the workflow separate when the privacy settings determine that local handling is required.
The main limitation is that Hybrid Compute is not a fully offline version of Perplexity. Cloud models still perform the web search, planning, and frontier-reasoning portions of supported tasks. Users who need all information to remain off external services should review each task carefully and avoid assuming that local processing applies to every part of an AI request.
How does Perplexity Hybrid Compute protect sensitive information?
Perplexity Hybrid Compute uses an on-device privacy gate that can decide whether information remains on the Mac, is masked before cloud processing, requires user approval, or cannot be used for an action. Perplexity says the classifier is designed to identify names, addresses, account numbers, credentials, payment-card numbers, government IDs, and secrets before the information leaves the device.
The privacy gate matters because private details often appear inside otherwise ordinary files and conversations. A document can contain a mailing address, a customer account number, or an access credential even when the user’s main request is simply to summarize or organize material. Perplexity’s approach is intended to inspect information locally before the system sends relevant portions of a task to a cloud model.
Perplexity describes the privacy gate as a protective control, not a guarantee that every sensitive detail will always be classified perfectly. Users should still avoid placing passwords, payment details, government identification numbers, and other highly sensitive material into AI workflows unless the local handling and permission settings are clear. The practical response is to review the model picker and privacy prompts before submitting a work or personal document.
Privacy controls have become more important as AI products retain more context across tasks. Readers comparing this approach with other AI services can also consider how shared-memory controls affect the information an assistant can use across conversations.
What Mac requirements does Perplexity Hybrid Compute need?
Perplexity Hybrid Compute requires an Apple-silicon Mac running macOS 15 or later with at least 24GB of unified memory. Perplexity set those requirements because the Mac must run a local AI model alongside the Perplexity app and the rest of the user’s workload. Macs that do not meet the memory requirement cannot use the feature as announced.
The 24GB unified-memory requirement is the most important compatibility limit for most Mac owners. Unified memory is the shared memory available to Apple silicon components, including the CPU and GPU. A Mac can run the Perplexity app without necessarily having enough memory for the local-model options used by Hybrid Compute.
| Requirement or option | Perplexity’s stated detail | Why it matters |
|---|---|---|
| Mac hardware | Apple-silicon Mac | Intel-based Macs are not listed as supported for Hybrid Compute. |
| Operating system | macOS 15 or later | The Mac needs a supported software version before the feature can run. |
| Memory | At least 24GB of unified memory | The local model requires substantial memory capacity. |
| Local-model options | Gemma 4 E4B, Qwen3.6 35B-A3B, or a Perplexity model | Users can choose a local model for the on-device portion of a task. |
Perplexity Hybrid Compute does not make an unsupported Mac eligible through a subscription upgrade alone. Check the Mac’s chip, macOS version, and unified-memory configuration before downloading a local model. The most sensible approach is to confirm compatibility first, then decide whether the local-processing benefit justifies the additional storage and computing demands.
How do you set up Perplexity Hybrid Compute in the Mac app?
Perplexity Hybrid Compute is set up by downloading a local model, selecting Hybrid in the model picker, and choosing the local and cloud models used for a task. Perplexity says the local-model download is available in one click, which reduces the setup work for users whose Macs meet the stated hardware and software requirements.
- Open the Perplexity Mac app on a supported Apple-silicon Mac.
- Download one of the available local-model options through the app.
- Select Hybrid in the model picker.
- Choose the local model and cloud model for the task.
- Review any permission request before sensitive information is sent to cloud processing.
Perplexity Hybrid Compute gives users a choice of Gemma 4 E4B, Qwen3.6 35B-A3B, or a Perplexity model for local use. The available options matter because the local model is responsible for the parts of a task that Perplexity intends to process on-device. Perplexity has not presented the feature as a single fixed model configuration, so users should confirm the selected local and cloud models before relying on a workflow for sensitive material.
Users should stop and contact Perplexity support if a local-model download fails, the Hybrid option does not appear on a Mac that meets the listed requirements, or the app repeatedly handles information in a way that does not match the selected privacy settings. Do not treat a failed setup as proof that a privacy control is active.
Can Perplexity Hybrid Compute process tasks started from an iPhone or iPad?
Perplexity Hybrid Compute can process sensitive work initiated remotely from an iPhone or iPad when a Mac remains powered on and running Perplexity. MacStories’ hands-on coverage describes that arrangement, which lets the Mac act as the local-processing device for a task started elsewhere.
The remote workflow matters because a phone or tablet does not need to perform the local model work itself if the supported Mac is available. A user can begin a task away from the desk while the Mac handles the sensitive local portion. The convenience depends on the Mac remaining on, running the Perplexity app, and available for the requested processing.
The main privacy consideration is that remote access does not eliminate the need to understand where each part of the task runs. Perplexity’s model design still assigns web search, planning, and frontier reasoning to cloud models. Users should check the privacy gate’s behavior and any approval request before assuming a document or detail will remain only on the Mac.
AI tools that use personal context can be useful for routine organization, but the same principle applies to features such as an assistant that can remember where personal items are stored. Users should decide which categories of personal information are appropriate for each service before enabling convenience features.
What controls do Enterprise administrators get with Hybrid Compute?
Perplexity Hybrid Compute gives Enterprise administrators organization-wide rules for three data categories: information that must remain on-device, information that can be masked for cloud use, and information that requires user approval before leaving a device. Perplexity announced those controls alongside the Mac launch, making the feature relevant to organizations that need consistent handling rules across employee devices.
The administrative controls matter because individual settings can produce inconsistent results when employees use different workflows. An organization can establish a default rule for sensitive categories rather than asking every employee to make a separate judgment during a task. The user-approval category also provides a point where a person can decide whether the cloud portion of a request is appropriate.
Enterprise rules do not remove the need for a company to define what counts as sensitive information in its own environment. Perplexity’s classifier is designed to recognize common categories such as credentials and payment-card numbers, but organizations may have additional confidential material, including internal documents or customer records. Administrators should map their internal data policies to the available on-device, masking, and approval settings before rolling out the feature broadly.
What is PII-TRACE, and why does it matter for Perplexity Hybrid Compute?
PII-TRACE is Perplexity’s 13-language benchmark for evaluating detection of recurring sensitive details across long conversations. Perplexity also introduced a compact 0.6-billion-parameter PII-Tracer model that the company says is intended to improve that detection. The benchmark and model support the broader goal of recognizing sensitive information before a hybrid workflow sends relevant content to cloud processing.
PII means personally identifiable information, which can include details that identify or help identify an individual. The long-conversation focus matters because sensitive information may not appear in one obvious message. A name in an early exchange, an account number later, and a location reference afterward can create a more sensitive combined context than any one detail alone.
Perplexity’s benchmark and model announcement does not establish that sensitive-data detection will be perfect in every language, document type, or conversation. The company’s claims describe the design and intended improvement, rather than an independent privacy guarantee. Users should continue to apply the same caution they would use with any AI assistant that receives private files or account-related information.
Should Mac users enable Perplexity Hybrid Compute?
Perplexity Hybrid Compute is most useful for eligible Mac users who want local handling for private files and sensitive details while still using cloud AI for web search and complex reasoning. The feature offers a clearer separation between local and cloud work than a workflow that sends an entire request to a remote model without a stated on-device privacy gate.
Perplexity Hybrid Compute is not necessary for every user. People who only use Perplexity for public web questions may not benefit enough from local-model setup to justify the memory requirement and configuration steps. People who handle highly sensitive data should also avoid treating the feature as a substitute for their organization’s security policies, legal obligations, or professional guidance.
The feature arrived as AI providers continue to add more capable tools and more persistent forms of context. Users should apply the same careful review to AI access and automation features, including browser-based systems with autonomous browser actions, because convenience changes how much information and control a service can receive.
Perplexity Hybrid Compute is a practical option for qualifying Mac owners, but the safest choice is to test the privacy controls with non-sensitive material first. Confirm that the local model is active, understand when the app requests approval, and keep credentials, payment information, and other high-risk data out of a workflow unless the handling path is clear.
FAQ
Is Perplexity Hybrid Compute available to free Perplexity users?
Perplexity Hybrid Compute is available to Perplexity Pro, Max, and Enterprise subscribers in the Perplexity Mac app. Perplexity’s September 1, 2026 announcement does not list free users among the supported subscription tiers.
Does Perplexity Hybrid Compute work on Intel Macs?
Perplexity Hybrid Compute requires an Apple-silicon Mac. Perplexity also requires macOS 15 or later and at least 24GB of unified memory, so an Intel Mac does not meet the announced hardware requirement.
Does Perplexity Hybrid Compute keep every part of a request local?
Perplexity Hybrid Compute does not keep every part of a request local because cloud models handle web search, planning, and frontier reasoning. The local model handles private files, sensitive information, and on-device actions when the privacy settings direct that handling.
What sensitive information can Perplexity Hybrid Compute identify?
Perplexity Hybrid Compute is designed to identify names, addresses, account numbers, credentials, payment-card numbers, government IDs, and secrets. Perplexity says the privacy gate can then keep that information local, mask it, refuse an action, or request permission before cloud transfer.
Can Enterprise administrators require data to stay on a Mac?
Perplexity Hybrid Compute lets Enterprise administrators set organization-wide rules for data that must remain on-device. Enterprise administrators can also define data that can be masked for cloud use or requires user approval before leaving a device.
