Perplexity Opens Waitlist for Always-On Local AI Agent on Mac
Perplexity’s new waitlist turns a spare Mac into a persistent local AI agent with approvals, logs, and a kill switch.
Perplexity opened a waitlist for Personal Computer on March 11, 2026, as part of a broader product expansion announcement. The product turns a spare Mac, specifically positioned around a Mac mini, into a persistent local AI agent with access to your files, apps, and sessions. For developers building AI agents, this matters because Perplexity is moving agent execution from cloud sandboxes onto user-owned hardware that stays online 24/7.
The Product
Perplexity describes Personal Computer as an always-on local extension of Perplexity Computer and the Comet Assistant. The waitlist page says it gives those systems “local access” to your Mac machine and can be controlled “from any device, anywhere.”
Three control features were disclosed publicly at launch:
- User approval for sensitive actions
- Full action logging
- A kill switch
Those are the concrete promises on record. As of March 15, 2026, Perplexity had not published a public technical architecture document, install guide, or local security whitepaper for Personal Computer.
Context in Perplexity’s Agent Roadmap
This launch was a product-line expansion, not a standalone announcement in isolation. Perplexity launched Computer on February 25, 2026 as a cloud-based agent platform, then added major upgrades on March 6, and on March 12 announced Computer for Enterprise. Personal Computer fits directly into that sequence.
The pattern is clear: cloud agent first, then model and tooling upgrades, then enterprise rollout, then a local-device version for users who want persistent access to machine state.
That matters because local machine state is where many high-value workflows actually live. Browser tabs, desktop apps, files in progress, local dev environments, signed-in sessions, and long-running tools are hard to replicate cleanly in a remote sandbox.
Release Timeline
| Date | Event | What changed |
|---|---|---|
| 2026-02-25 | Perplexity launches Computer | Cloud-based agent system for Max subscribers, with 19-model orchestration and 10,000 monthly credits plus 20,000 bonus credits |
| 2026-03-06 | Perplexity upgrades Computer | Adds Custom Skills, Model Council, Voice Mode, and GPT-5.3-Codex coding subagent |
| 2026-03-11 | Perplexity opens Personal Computer waitlist | Announces always-on local Mac-based agent access |
| 2026-03-12 | Perplexity launches Computer for Enterprise | Cloud product becomes enterprise-available |
This sequencing suggests Perplexity is treating local and cloud agents as one stack with two execution environments.
Known Technical Details
Perplexity disclosed far more detail for Computer than for Personal Computer, but the launch context gives useful clues.
Perplexity said Computer orchestrates 19 models in parallel. Its published material named Claude Opus 4.6 as the main reasoning engine, with other subagents and specialist models including Gemini, Nano Banana, Veo 3.1, Grok, and ChatGPT 5.2, while describing the system as model-agnostic. On March 6, Perplexity added Model Council, which runs GPT-5.4, Claude Opus 4.6, and Gemini 3.1 Pro in parallel and synthesizes outputs.
There is no public statement confirming that the same orchestration runs identically on Personal Computer. But the launch language strongly suggests the local Mac device is an execution surface for the broader Computer system, not a completely separate local-only model runtime.
That distinction matters. “Local AI agent” here does not necessarily mean all model inference happens on the Mac. It means the agent has local presence and machine access. The planning and reasoning stack may still rely heavily on cloud models.
Local access changes the design tradeoffs
Perplexity’s cloud Computer product is framed around isolated environments, connectors, and sandboxed execution. Personal Computer shifts the emphasis to persistence and direct machine access.
Here is the practical tradeoff developers should pay attention to:
| Dimension | Cloud Computer | Personal Computer |
|---|---|---|
| Execution environment | Perplexity-managed cloud sandbox | User-owned Mac on local network |
| Access to local files/apps | Indirect, connector-based or sandbox-limited | Direct access to files, apps, sessions |
| Persistence | Task/session based | 24/7 always-on |
| Security controls publicly disclosed | Enterprise claims include SOC 2 Type II, SAML SSO, audit logs, admin controls | Publicly disclosed controls are approval, action logs, kill switch |
| Public documentation depth | Relatively detailed | Sparse, waitlist-level only |
If you build agentic software, the core takeaway is that persistent local presence unlocks workflows that browser-only or sandbox-only agents struggle with. It also expands the blast radius of mistakes.
Mac Mini as Dedicated Agent Host
Perplexity specifically positioned Personal Computer around the Mac mini. That is a meaningful product decision.
A spare Mac mini is cheap to dedicate, quiet, power-efficient, and stable enough to leave on continuously. It gives Perplexity a controlled hardware target compared with “install this on any desktop.” For agent systems, that reduces variability in permissions, networking, and session persistence.
For developers, this looks closer to a home lab appliance than a consumer app. A dedicated always-on Mac becomes the trust anchor for your agent, similar to how teams treat a build machine, headless browser box, or internal automation host.
Security Controls and Open Questions
Perplexity’s messaging around Personal Computer focused on auditability and human control. The approval, logging, and reversibility controls position it as a more secure alternative to systems like OpenClaw.
That is directionally sensible. Local agents need strong intervention controls because they can operate with authenticated sessions, filesystem access, and long-lived app state. A kill switch and action logs are foundational features, not premium extras.
But the public record still has gaps:
- No public security whitepaper for Personal Computer
- No public explanation of whether it runs as a daemon, app, VM, container, or browser bridge
- No published permission model
- No public detail on network exposure
- No benchmarked failure or rollback behavior
For engineering teams, this means the launch is strategically important, while the implementation risk remains hard to quantify.
This is part of a broader Perplexity agent push
Personal Computer landed days after Perplexity expanded Computer with new agent features and the same day it announced Computer for Enterprise. Perplexity also cited internal usage numbers for Computer from a study of more than 16,000 queries, claiming $1.6 million in labor cost savings and the equivalent of 3.25 years of work completed in four weeks.
Those figures are about the broader Computer platform, not Personal Computer specifically. They still matter because they show how Perplexity is pitching the category: not as chat, but as a digital worker platform with measurable output.
The local Mac version extends that pitch into workflows where cloud isolation is too limiting.
Open Questions
Several missing details will determine whether Personal Computer becomes a serious platform or a niche demo:
| Unknown | Why it matters |
|---|---|
| Pricing | Determines whether this is a consumer sidecar or a serious automation product |
| Hardware requirements | Affects whether M-series Macs are required and what concurrency is realistic |
| Installation model | Determines security posture, update strategy, and manageability |
| Local vs cloud execution split | Affects privacy, latency, and offline behavior |
| API or extensibility surface | Determines whether developers can integrate tools, structured output, or custom workflows |
| Failure handling | Critical for trust in unattended operation |
If Perplexity exposes a developer surface here, the product could become a local runtime for agent tasks that need desktop context, much like how RAG systems became practical once teams could attach retrieval to general-purpose LLMs. If it stays closed and consumer-oriented, its impact will be narrower.
Competitive context
Perplexity is entering a crowded area that includes cloud computer-use agents, browser automation tools, and local-first desktop operators. Its visible differentiation is the combination of:
- Persistent local machine presence
- Multi-model orchestration from the Computer stack
- Approval and audit controls
- Remote access from other devices
That combination is stronger for real-world desktop work than single-step browser agents. It is also harder to secure and harder to support.
The important technical shift is not simply “agent on your computer.” The shift is agent with durable machine state. That includes open sessions, local apps, background processes, and ongoing context across time. For autonomous workflows, that matters more than one-shot tool calling or a larger context window.
Implications for Local AI Agents
Perplexity’s March 11 launch gives a clear signal about where local agents are heading on Apple hardware. The target is not a chatbot that happens to run on a laptop. The target is a dedicated always-on host that an agent can use as its working environment.
If you build Mac automation, coding assistants, or desktop operators, expect the architecture to move toward three layers:
- Cloud reasoning and orchestration
- Local execution and state access
- Human approval, audit, and interrupt controls
That architecture is more practical than trying to force every workflow into either a pure cloud sandbox or a fully on-device model.
If your application depends on desktop context, signed-in apps, or persistent sessions, test your design against a dedicated-host model now. Treat approval gates, logs, and hard-stop controls as first-class product requirements before you optimize model choice or tokenization.
Get Insanely Good at AI
The book for developers who want to understand how AI actually works. LLMs, prompt engineering, RAG, AI agents, and production systems.
Keep Reading
Multi-Agent Systems Explained: When One Agent Isn't Enough
Multi-agent systems use specialized AI agents working together on complex tasks. Here's how they work, the main architecture patterns, and when they're worth the complexity.
AI Agents vs Chatbots: What's the Difference?
Not every AI chatbot is an agent, and not every task needs one. Here's the real distinction between agents and chatbots, the spectrum between them, and when each makes sense.
What Are AI Agents and How Do They Work?
AI agents can plan, use tools, and take action autonomously. Here's what they are, how they work under the hood, and what separates useful agents from overhyped demos.
Anthropic Makes Claude's 1M Token Context Generally Available
Anthropic made 1M-token context GA for Claude 4.6, removing long-context premiums and boosting throughput for large code and agent tasks.
What Are Agent Skills and Why They Matter
Agent skills are portable packages of instructions that extend AI coding agents. Here's what they are, how they work, and why the open standard changes how developers work with AI tools.