Perplexity has introduced a fully local version of its Computer AI agent for Apple Silicon Macs, allowing supported tasks to run directly on a user’s computer instead of relying on cloud-based AI processing. Announced on October 20, 2026, the feature is available to Perplexity Pro, Max and Enterprise subscribers through the company’s Mac application.
Source verification, 21 October 2026: Perplexity’s official local Computer announcement. Fully local tasks differ from hybrid tasks; cloud assistance requires permission.
The update means the AI model, task planning, reasoning and execution can all run on-device. With user permission, Computer can work across local files and native Mac applications, including Notes, Messages, Mail and Numbers, without sending the contents of those files to remote AI servers during a fully local task.
Perplexity is launching the feature with two local model options: Qwen 3.6 and PPLX Pro, a Perplexity-trained version of Qwen. The company recommends a Mac with at least 36GB of unified memory for running Computer locally.
The release follows Perplexity’s earlier Personal Computer and Hybrid Compute offerings, which brought AI assistance to local Mac files and combined on-device processing with cloud-based reasoning. The new mode goes further by allowing the complete supported workflow to run locally, while leaving users in control of when a task may use cloud services.
The launch highlights a growing shift in AI computing: instead of sending every request and its associated data to remote servers, increasingly capable personal computers can handle more complex AI workflows on the device itself.
Key takeaways
- Fully local AI: Computer’s model, planning, reasoning and task execution can run on supported Apple Silicon Macs.
- Available to paid subscribers: The feature is available to Perplexity Pro, Max and Enterprise users.
- 36GB memory recommended: Perplexity recommends at least 36GB of unified memory for local operation.
- Two model options: Users can choose Qwen 3.6 or PPLX Pro for local processing.
- Works across Mac apps: With permission, Computer can work with local files and applications such as Notes, Messages, Mail and Numbers.
- Cloud access remains optional: Users can authorize cloud processing for current information, browser use, connected apps or more advanced reasoning.
- Privacy and control: Local execution can keep sensitive files on the Mac, although it does not eliminate every security risk or guarantee that every workflow works offline.
What is Perplexity Computer?
Perplexity Computer is an AI agent designed to carry out multi-step tasks rather than simply answer questions in a chat window.
A conventional chatbot might explain how to organise a project or draft an email. An AI agent can go further by gathering relevant information, working with files, updating documents and preparing outputs across several applications, subject to the permissions available to it.
For example, a user might ask Computer to review recent messages, identify outstanding deadlines, compare them with a project checklist and prepare a summary in Notes. Another workflow could involve analysing financial records in a spreadsheet and preparing a presentation for review.
The objective is to reduce the amount of manual work required to move between applications and complete related tasks.
Perplexity’s new local mode changes where these activities take place. Instead of relying on remote infrastructure for the entire AI workflow, Computer can use the Mac’s own processing resources to handle the model and coordinate the task.
This matters particularly when a task involves confidential documents, personal messages, internal financial information or other files that a user would prefer to keep on their own machine.
What does fully local AI mean?
In a cloud-based AI workflow, information is sent to remote servers where the model processes the request and generates a response. Depending on the application, the cloud service may also coordinate tools and manage parts of the task.
With Perplexity’s fully local Computer mode, the model and the agent’s main task workflow run on the Mac itself.
| Capability | Fully local Computer mode |
|---|---|
| AI model inference | Runs on the Mac |
| Task planning and coordination | Runs on the Mac |
| Reasoning for supported tasks | Runs on the Mac |
| Processing local files | On-device, with permission |
| Working with supported native apps | Available with permission |
| Code and tool execution | Uses isolated sandbox environments |
| Current web research | May require authorised cloud access |
| Cloud-connected applications | May require authorised cloud access |
| Access to advanced cloud models | Available when the user authorises escalation |
The distinction is important because local processing does not mean every possible task is automatically offline.
A task involving files already stored on the Mac may be handled locally. A task requiring live web information, a cloud-connected service or a more capable remote model may need an internet connection and authorised cloud access.
Users can therefore choose between keeping a workflow on the device and allowing selected parts of a task to use cloud resources.
Which AI models power Perplexity Computer locally?
Perplexity says the new mode launches with two options: Qwen 3.6 and PPLX Pro.
Qwen 3.6 is the underlying model family used in the local setup, while PPLX Pro is a Perplexity-trained version of Qwen intended for the company’s Computer workflow.
Running these models locally means the Mac performs the model’s inference using its own hardware. The available performance will depend on the computer’s memory, processor, model configuration and the complexity of the task.
Perplexity has not provided a comprehensive, independently verified benchmark showing how complete Computer workflows perform across every supported Mac configuration. As a result, the availability of the feature should not be interpreted as a guarantee that every Mac will complete complex tasks at the same speed.
The launch is nevertheless significant because it combines local model execution with an AI agent capable of coordinating multi-step work, rather than offering only a standalone local chatbot.
Why does Perplexity recommend 36GB of unified memory?
Perplexity recommends at least 36GB of unified memory for Computer’s fully local mode.
Unified memory is shared by the CPU, GPU and other parts of Apple Silicon systems. Local AI models require memory to hold model parameters and intermediate data while processing requests. The operating system and other running applications also need memory.
For an AI agent, available memory can affect more than the speed of generating text. The system may also need resources for planning, tool use, document processing and other operations that occur during a task.
A Mac with insufficient memory may struggle to accommodate the local model and the rest of the workflow comfortably. Perplexity’s recommendation is therefore an important practical consideration for people deciding whether to use the feature.
Which Macs are suitable?
The key requirement is a supported Apple Silicon Mac with enough memory for the local workload. Perplexity highlights newer Mac mini systems, including configurations with M6 and M5 Pro chips, among the machines that can run the feature.
The memory configuration matters as much as the chip name. Some Macs offer configurations that fall below the recommended 36GB, while higher-memory versions may meet or exceed it.
| Consideration | What users should check |
|---|---|
| Processor | The Mac must use supported Apple Silicon hardware |
| Unified memory | At least 36GB is recommended |
| Application | The latest Perplexity Mac app is needed |
| Model setup | A supported local model must be downloaded and configured |
| Subscription | Pro, Max or Enterprise access is required |
| Cloud-dependent workflows | Internet access and authorisation may still be necessary |
Perplexity describes 36GB as a recommendation, so it should not automatically be interpreted as a strict universal minimum for every possible configuration. Users should check the current in-app requirements before making hardware decisions.
What can Perplexity Computer do with local Mac apps?
The new mode is designed to let Computer work with files and supported native applications, provided users grant the required permissions.
1. Review emails and messages
Computer can examine selected information in Mail and Messages, identify requests or deadlines and help turn them into an organised action list.
For instance, a user could ask it to identify outstanding work from recent conversations and prepare a summary in Notes.
2. Organise tasks and documents
The agent can use local project files and to-do lists to prepare summaries, update documents or organise information across supported workflows.
This could help professionals who manage multiple projects and need to consolidate information from different sources.
3. Analyse spreadsheets and financial records
Computer can work with local financial information, such as downloaded market data and private portfolio records, to prepare an analysis or update a workbook.
The ability to process private information locally may be particularly valuable to people who do not want to send confidential records to an external AI service.
However, financial analysis generated by an AI agent should still be checked for errors before being used to make investment, accounting or business decisions.
4. Prepare presentations and business documents
The agent can help transform information from documents and spreadsheets into presentation material, reports or draft communications.
Users can review the results before using or sharing them. This is important because an agent may misunderstand a request, overlook information or produce an inaccurate summary.
The common theme across these examples is that Computer is intended to move beyond generating text to coordinating tasks across applications.
How local execution improves privacy
One of Perplexity’s main arguments for fully local processing is that sensitive information can remain on the user’s device.
In a cloud-based workflow, a request may require documents or other context to be transmitted to a remote service. In a fully local workflow, the model can process local files on the Mac without sending their contents to remote AI servers.
This can be useful for several categories of work:
- Personal information: Messages, emails and personal documents.
- Business records: Internal budgets, client briefs and operational documents.
- Financial information: Private portfolio holdings and downloaded financial records.
- Project material: Drafts, plans and other files that have not been shared publicly.
Perplexity says code and tool execution take place in isolated sandbox environments with controlled access to files and supported applications.
These safeguards can help limit the scope of what the agent can access. Nevertheless, local execution does not make an AI agent inherently risk-free. A system that has permission to read or modify files can still make mistakes or respond inappropriately to misleading instructions embedded in documents.
Users should grant only the permissions necessary for the tasks they want to perform and review important actions before allowing an agent to send communications, alter files or share information.
Can Perplexity Computer work without an internet connection?
Perplexity says the fully local mode can run tasks entirely on the Mac, including workflows that do not require cloud services.
However, the practical answer depends on what a task needs.
A task involving local files and a downloaded model may be able to run without internet access. A task requiring live news, current market information, a website or a cloud-connected application may need a connection.
Users can also authorise Computer to escalate a task to cloud-based services. Perplexity says this can provide access to current information, browser use, connected apps and more than 15 frontier AI models for advanced reasoning.
The cloud escalation option is useful when the local model cannot handle a request effectively, but it changes the data-processing arrangement. Users who want to keep a task fully local should make sure cloud access is not authorised for that workflow.
Fully local Computer vs Hybrid Compute
The new launch follows Perplexity’s Hybrid Compute feature, which combines local processing with cloud-based AI.
The two modes are designed for different needs. Fully local Computer prioritises on-device execution, while Hybrid Compute allows the system to use cloud models for demanding reasoning and local models for sensitive information.
| Feature | Fully local Computer | Hybrid Compute |
|---|---|---|
| Main task workflow | Runs on the Mac | Combines local and cloud processing |
| Local file processing | On-device | Can remain on-device |
| Cloud reasoning | Requires user-authorised escalation | Part of the hybrid workflow |
| Access to current online information | Requires authorised online access | Available through cloud components |
| Privacy approach | Keeps the supported workflow on-device | Can keep sensitive steps local while using cloud capabilities elsewhere |
| Recommended memory | 36GB or more | Perplexity says it supports Macs with at least 24GB of unified memory |
| Best suited for | Tasks that should remain local | Tasks needing a mix of privacy and cloud capability |
The key difference is not simply speed. It is where information is processed and how much control the user wants over the use of cloud resources.
Hybrid Compute may be more practical for users whose Macs do not meet the recommended memory level for fully local operation or who frequently need online research. Fully local mode may be more appropriate when the task can be completed with the available local model and the user prioritises keeping information on the device.
How to enable fully local Perplexity Computer on Mac
Perplexity says users can get started through its Mac app.
- Download and open the latest Perplexity Mac application.
- Follow the in-app instructions to download and set up a supported local model.
- Open Computer and select the fully local mode.
- Give the agent permission to access the specific files or applications required for the task.
- Review the task and its output, and decide whether any cloud access is appropriate.
The feature is available to Perplexity Pro, Max and Enterprise subscribers on supported Apple Silicon Macs.
Users should confirm that their memory configuration is suitable before relying on the feature for demanding workflows. They should also check the app’s current permissions and model options, as availability and setup requirements may change.
Why this launch matters for the future of AI agents
AI assistants are increasingly being designed to complete work rather than simply respond to prompts. That shift raises new questions about privacy, computing costs, reliability and user control.
Cloud-based AI gives applications access to large models and remote computing resources. However, sending sensitive data to remote systems can be a concern for individuals and organisations.
Local AI offers a different approach. As computers become more capable and models become more efficient, more of the work involved in an AI task can be performed on the device itself.
Perplexity’s launch combines three elements: a local model, an agent that coordinates a task and tools that can interact with files and supported applications. The combination is more significant than running a model locally in isolation because it allows the system to work through multi-step tasks.
There are still trade-offs. Local systems depend on available memory and processing power, while cloud models may offer stronger capabilities for some complex tasks. Local execution also does not eliminate the need for careful permission management and human oversight.
The likely direction is not that every AI task will become local. Instead, users may increasingly choose where different parts of a workflow run, depending on the sensitivity of the information, the model’s capabilities and the hardware available.
What this means for businesses and professionals
For professionals who work with sensitive documents, fully local AI could provide a way to use an agent without automatically transmitting those files to a remote AI service.
A consultant could organise client documents. An analyst could work with local financial records. A small-business owner could prepare summaries from internal files. In each case, the ability to process data locally may help address privacy concerns.
Enterprise users will still need to consider access controls, organisational policies, recordkeeping and the risks associated with automated actions. The fact that a task runs locally does not automatically make it compliant with every company’s security requirements or applicable regulations.
Businesses should also evaluate whether the local model is accurate enough for their workloads and whether the time saved justifies the computing resources required.
The launch provides another option for organisations that want to experiment with AI agents while maintaining greater control over where information is processed.
Frequently asked questions
1. What is Perplexity Computer’s fully local mode?
It is a version of Perplexity’s AI agent that runs its model, planning, reasoning and task execution on supported Apple Silicon Macs rather than relying on cloud processing for the entire workflow.
2. Who can use Perplexity Computer locally on Mac?
The feature is available to Perplexity Pro, Max and Enterprise subscribers through the Perplexity Mac app.
3. How much memory does Perplexity Computer require?
Perplexity recommends at least 36GB of unified memory for fully local operation. Users should check their specific Mac configuration and the latest in-app requirements.
4. Which AI models are available locally?
Perplexity says the feature launches with Qwen 3.6 and PPLX Pro, a Perplexity-trained version of Qwen.
5. Can Perplexity Computer access private files?
Yes, with user permission, it can work with local files and supported Mac applications such as Notes, Messages, Mail and Numbers. Local processing allows supported tasks to be completed without sending the contents of those files to remote AI servers.
6. Does fully local Computer work offline?
Tasks that use only the downloaded local model and local resources may run without an internet connection. Current web research, cloud-connected apps and other online services may require internet access.
7. Is fully local Computer more private than cloud-based AI?
Local processing can reduce the need to send sensitive information to remote AI servers. However, privacy also depends on permissions, application security and how the agent handles files and instructions.
8. What is the difference between fully local Computer and Hybrid Compute?
Fully local Computer runs the supported task workflow on the Mac. Hybrid Compute combines local processing with cloud-based models and services, allowing users to benefit from remote reasoning while keeping selected sensitive work local.
9. Can Perplexity Computer send emails or modify documents automatically?
Computer can work across supported applications and files when granted permission. Users should review important outputs and actions, particularly when they involve sending communications, changing records or sharing sensitive information.
10. Does running AI locally eliminate cloud costs?
It can reduce the need to use cloud AI for tasks that the local model can complete. However, cloud services may still be used when authorised, and the Mac itself consumes computing resources. The overall cost depends on the user’s subscription, usage and hardware.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
