Perplexity Computer is the execution layer inside Perplexity for research and knowledge-work tasks. This review focuses on the September 2026 Effort Mode, GPT-6 Astra, local-first Portable Computer, Hybrid Compute on Mac, and Skills Marketplace announcements. We did not independently test the product or claim that local execution is automatically private.
What Perplexity Computer does
Computer can plan and carry out multi-step tasks rather than returning only a cited answer. Effort Mode adds Light, Standard, High, and Ultra levels so a user can choose a rough cost-and-depth target while Perplexity selects the model and reasoning level. Eligible Pro and Max subscribers can use GPT-6 Astra for Computer tasks and supporting agents, subject to plan and data-retention eligibility. Side Chat provides a read-only place to ask a question alongside a running task without changing the main task.
Perplexity also describes Portable Computer for supported NVIDIA DGX Spark and Windows or Linux PCs with NVIDIA RTX GPUs and at least 24 GB of VRAM. The orchestrator, models, tools, local search, and queue can run on the device; cloud escalation requires approval. Hybrid Compute on Mac splits work between local models and the cloud, with local checks for private data and an option to continue from an iPhone. These hardware and plan conditions are important product limits, not footnotes.
Best fit and not fit
Best fit: research teams, analysts, and professionals who want a visible effort dial, sourced web research, and a local-first path for sensitive files on supported hardware.
Not fit: buyers without the required GPU or memory, teams that need a fully self-hosted stack, or workflows that cannot tolerate a cloud escalation or connected-app permission. “Local-first” does not mean that every request, update, or connector stays on the device.
Our editorial decision rule
Start with a low-risk task and compare Light, Standard, and High on completion quality, runtime, credits, and correction effort. For Portable Computer or Hybrid Compute, document the hardware, model, update path, cloud-escalation setting, and which files remain local. For Skills Marketplace, inspect the skill owner, requested permissions, package contents, and update process before installing it for a team. Choose Computer when the execution loop saves more review time than a cited search; choose ordinary Search or a controlled local workflow when you cannot govern agent actions.
Evidence and limitations
The official Perplexity changelog supports the feature, plan, hardware, and availability descriptions. It does not establish accuracy, latency, privacy, or reliability for every task. We have not independently tested Effort Mode, Portable Computer, Hybrid Compute, GPT-6 Astra access, or Skills Marketplace governance. Read the Perplexity profile, compare Perplexity vs Gemini, and use the AI privacy checklist before connecting confidential material. Re-check current plan rules and cloud-escalation controls before adoption.nn
What Perplexity says runs locally
On October 20, Perplexity announced a fully local Computer mode in its Mac app. The company says the model, harness, and task execution can run on device, and that—with user permission—the feature can work across local files and native apps such as Notes and Mail without sending that personal data to the cloud. Perplexity says the launch includes Qwen 3.6 and PPLX Pro as local model options. This is a materially different privacy and hardware proposition from the earlier Hybrid Compute design, where frontier reasoning may run in the cloud while local models process private information on the Mac.
Perplexity lists availability for Pro, Max, and Enterprise subscribers on supported Apple-silicon Macs and recommends 36 GB or more of memory. These are vendor-stated requirements and availability details; confirm the latest app, device, and account eligibility before planning a rollout. “Fully local” should be read narrowly as the announced task-execution mode, not as proof that every integration, account event, update, or optional cloud feature is offline.
Who should consider it—and what to verify first
This mode may suit people handling sensitive local documents who want an on-device workflow and have compatible hardware. It is less suitable when a team needs centrally managed endpoint controls, a guaranteed offline environment, or a formal data-processing commitment that has not been checked against the applicable enterprise terms. Before use, inspect the app’s permission prompts, test with non-sensitive files, disable unnecessary connectors, and confirm whether each task stays local when it invokes a native app or external service.
The source is Perplexity’s product announcement, not an independent privacy audit. We cannot verify the implementation or infer that all data paths are local from marketing language alone. For model, credit, and connector changes in the same release window, see the Perplexity review update. Next review: November 11, 2026.