Perplexity brings Portable Computer to AMD Ryzen AI Max PCs

Perplexity has added Windows support for AMD Ryzen AI Max processors to Portable Computer, its subscription-based local agent runtime. Eligible AMD systems can now run models, orchestration, and tool calls on-device, expanding a product that previously required NVIDIA hardware.

The addition gives laptop and workstation buyers another hardware option for private AI workflows. Model inference and eligible workflow data can remain on the PC, while cloud escalation requires approval. Actions involving external services, such as posting to Slack, still send the approved data to that service.

Inside the local agent stack

Portable Computer is the on-device version of Perplexity Computer, the company’s multistep agent. An agent harness manages the workflow around a model: it interprets a request, divides the work into steps, selects tools, tracks progress, and handles results.

Portable Computer runs that harness, including its planner, orchestrator, and tool router, on a compatible Windows PC. Supported local models include PPLX 27B and Qwen 27B, with Nemotron support planned. Perplexity has not announced a release date for Nemotron.

Cloud models remain available for tasks that need additional reasoning or current web data. The software identifies those steps and requests permission before sending task information to Perplexity’s servers, creating a visible boundary between local and hybrid execution.

Why Ryzen AI Max fits

Ryzen AI Max combines CPU cores, integrated Radeon graphics, and unified memory in one package. The CPU and GPU share the same memory pool, allowing high-memory configurations to reserve enough capacity for models that would exceed the dedicated graphics memory found in many laptops.

Perplexity’s stated hardware requirements are:

A 27-billion-parameter model would require about 54 GB for 16-bit weights alone, so deployment within a 24 GB graphics-memory target depends on lower-precision, quantized weights. Context caches, tool processes, and other applications also consume memory, which means context capacity and performance will vary by configuration.

The AMD path therefore depends on usable GPU-addressable memory, not the processor badge alone. Buyers must check the system’s total memory, firmware allocation options, and selected configuration. Perplexity cites HP’s ZBook Ultra G3a with Ryzen AI Max PRO 400-series processors as one compatible product family, subject to configuration.

Workflows, connectors, and schedules

Portable Computer connects with Gmail, Outlook, Slack, GitHub, and files stored on the PC. Those integrations support workflows such as:

• Reviewing the latest pull requests in a repository and posting a summary to Slack.
• Running a pull-request review each day at 9 a.m.
• Producing a weekly digest from a local folder.
• Summarizing a directory of PDFs without usage-based cloud inference charges.

Scheduled jobs can run at recurring intervals while the host machine remains available. Work completed entirely on-device does not consume Perplexity Computer credits, although access to Portable Computer still requires an eligible subscription.

Permissions at the device boundary

Portable Computer executes local tasks inside a sandbox and brokers access to files, applications, and connectors. The broker acts as an intermediary, limiting direct access and applying the permissions granted to each tool.

A workflow that summarizes tax documents and posts the result to Slack can read and process the files locally, then display an approval request before contacting Slack. The outbound action sends the approved result to Slack under that connector’s permissions.

Enterprise evaluations should examine the controls around the sandbox and connected services, including:

• OAuth scopes for each connector.
• Local file and directory permissions.
• Logs for cloud escalation and outbound tool calls.
• Process isolation and sandbox boundaries.
• Model update signing and version controls.
• Administrative restrictions for tools, schedules, and connectors.

Subscription and setup

Portable Computer is available to Pro and Max subscribers across individual and enterprise plans. The subscription covers the agent harness, orchestration layer, connectors, scheduling, updates, and optional access to cloud models. Local inference uses the customer’s hardware and does not incur per-token charges for locally completed work.

Windows setup begins in the Perplexity desktop application. Users select a supported local model, download it, and configure the required connectors and permissions. On Ryzen AI Max systems, the machine must expose at least 24 GB of graphics-addressable unified memory to meet Perplexity’s requirement.

AMD widens the local-agent market

Adding Ryzen AI Max gives manufacturers and enterprise buyers an alternative to NVIDIA for high-memory local inference. It also extends Portable Computer to integrated systems designed for mobile workstations, where a discrete GPU can increase size, power use, and cost.

The change has three practical effects:

1. PC makers using AMD’s high-memory processors gain a packaged agent runtime for local workflows.
2. Enterprise procurement teams gain a second hardware path for on-device AI deployments.
3. Agent developers gain another target for optimization beyond NVIDIA’s CUDA-centered ecosystem.

Compatible systems can now handle scheduled document processing, repository analysis, and other sensitive or high-volume work with a 27B local model. Cloud models remain available when a task exceeds local capabilities, with approval controlling when information leaves the device.