Whether you're looking for an angel investor, a growth advisor, or just want to connect — I'm always open to great ideas.
Get in TouchPerplexity launches Personal Computer at Ask 2026 — software that turns a Mac mini into an always-on AI agent connected to Gmail, Slack, GitHub, and Salesforce, with your choice of Claude, Gemini, or Grok as the brain.
AI, startups & growth insights. No spam.
TL;DR: At its inaugural Ask 2026 developer conference, Perplexity unveiled Personal Computer — software that runs continuously on a user-supplied Mac mini and gives AI agents persistent, local access to files, apps, Gmail, Slack, GitHub, Notion, and Salesforce. Unlike Perplexity Computer (the cloud-only agent), Personal Computer lives on your hardware, executing tasks around the clock without you being present. It costs $200/month for Perplexity Max subscribers, launches via waitlist, and supports multi-model deployment across Claude, Gemini, and Grok.
Perplexity already had a cloud-based AI agent product called Perplexity Computer — a multi-model orchestration system that runs research, coding, and project-completion tasks entirely inside Perplexity's secure cloud sandbox. Computer was powerful, but it was disconnected from your local machine. It could not open your files, read your calendar, or interact with desktop apps.
Personal Computer changes that.
Personal Computer is software you install on a Mac mini sitting on your desk (or anywhere with power and internet). Once installed, it runs continuously — 24 hours a day, 7 days a week — and acts as a local bridge between Perplexity's cloud AI and your actual computing environment. The agent has access to local files and applications, not just cloud APIs.
The distinction matters enormously. Cloud-only agents like the original Computer work on data you explicitly hand them. Personal Computer works on your data as it naturally lives — emails as they arrive in Gmail, commits as they land in GitHub, tickets as they move through your Salesforce pipeline. You do not have to copy-paste anything. The agent monitors, reacts, and executes.
Perplexity describes the goal simply: you hand the system an objective, not a list of commands. The agent figures out which files to open, which apps to invoke, and how to get it done. Users in early access reported describing broad goals — things like "create an interactive educational guide" or "compile a competitive analysis from our Notion workspace" — and the system independently navigated to the right data and produced finished outputs.
Personal Computer was announced on March 11, 2026 at Ask — Perplexity's inaugural developer conference, held inside a former church in San Francisco's North Beach neighborhood. The venue choice felt intentional: reverent, slightly dramatic, and unmistakably San Francisco.
CEO Aravind Srinivas used the keynote to articulate a vision that has been building in Perplexity's public communications for months.
"A traditional operating system takes instructions; an AI operating system takes objectives." — Aravind Srinivas, Ask 2026
That one sentence captures the philosophical shift Perplexity is betting on. Every operating system since CP/M has worked the same way: you tell the computer exactly what to do, one command at a time. Srinivas argues that AI unlocks a different paradigm — you describe what you want achieved, and the system reasons about how to achieve it.
Personal Computer is the most concrete expression of that thesis so far. It is not a chatbot you interact with. It is an agent that runs whether you are watching or not.
The conference also served as the venue for Perplexity's expanded enterprise Computer announcements, an update to Perplexity's Hey feature on Samsung devices, and a broader signal that Perplexity is shifting from "AI search" to "AI infrastructure."
Sources covering the event included Axios, 9to5Mac, Neowin, Apple Insider, and Digital Trends.
The Mac mini was not a random hardware choice. Several factors make it the natural fit for an always-on local AI agent:
Cost. A base Mac mini starts around $599 — cheap enough to dedicate entirely to background AI work without taking your main machine out of commission.
Apple Silicon. The M-series chips have fast unified memory architecture and efficient neural engine acceleration. Running local inference workloads or coordinating API calls to cloud models is fast and energy-efficient on Apple Silicon in a way that x86 laptops running at full clock speed cannot match for sustained workloads.
Silent, small form factor. The Mac mini runs fanlessly under light loads. You can put it in a closet, on a shelf, or behind a monitor and forget it is there. It is not a workstation anchoring your desk.
macOS ecosystem. macOS has mature, scriptable app automation via AppleScript and Shortcuts, robust file system permissions, and broad enterprise software compatibility. Perplexity's agent can hook into those affordances in ways that would require more custom work on Windows or Linux.
Personal Computer is Mac-only at launch. The Macworld coverage noted that Perplexity has not ruled out Windows support but has made no commitments. The Mac-first strategy lets Perplexity ship a polished, well-integrated product before expanding to a platform with more hardware variability.
You do not have to buy a Mac mini specifically — Personal Computer will run on any Mac. But the Mac mini is the natural recommendation because it is cheap, always-on-capable, and does not require you to leave a more expensive laptop running around the clock.
The power of Personal Computer is not the AI itself — it is the integrations. Out of the box, the system connects to:
| Integration | What the agent can do |
|---|---|
| Gmail | Monitor inbox, draft replies, extract action items, archive threads |
| Slack | Watch channels for triggers, post updates, summarize threads |
| GitHub | Review PRs, track issues, flag build failures, generate summaries |
| Notion | Read and write pages, organize projects, cross-reference docs |
| Salesforce | Monitor pipeline changes, update records, generate reports |
These are not read-only connections. The agent can write: it can draft and queue emails, post messages, create GitHub issues, update Notion databases, and edit Salesforce records. This is why the safety architecture described below exists — write access to production systems demands careful guardrails.
The integration model mirrors what enterprise workflow tools like Zapier or Make have long offered, except the logic layer is a reasoning AI rather than a rigid if-then rule. Instead of "when Slack message contains X, create Jira ticket," you say "monitor our #engineering-alerts channel and open GitHub issues for anything that looks like a production incident."
Perplexity's enterprise Computer product connects to over 400 platforms including Snowflake and HubSpot. The initial Personal Computer launch ships with the core five integrations listed above, with more expected as the product moves out of waitlist phase.
Personal Computer does not lock you into a single AI model. Subscribers can configure which frontier models the agent deploys, choosing from:
This continues Perplexity's multi-model orchestration philosophy from the original Computer product, which coordinates 19 AI models including Claude Opus 4.6, GPT-5.2, Gemini, and Grok. The cloud product routes each subtask to the model best suited for it. Personal Computer brings a user-configurable version of that same approach to the local context.
You can deploy multiple models simultaneously. A practical setup might use Claude for reasoning and drafting, Gemini for research and Google Workspace tasks, and Grok for quick, time-sensitive lookups. Perplexity handles the orchestration; you set the policies.
This model-agnostic approach is deliberate strategy. Srinivas has said publicly that Perplexity's differentiation is orchestration — the harness that deploys models intelligently — not any single model. Locking users into one model would undercut that thesis and expose Perplexity to competitive risk if one provider's quality slips.
Giving an AI agent persistent access to your email, Slack, and code repositories is a significant trust decision. Perplexity has built a multi-layer safety architecture to make that trust defensible.
confirm_action protocol. Before executing sensitive operations — sending emails, making purchases, deleting files, modifying records — the agent surfaces a confirmation request. You see exactly what action is being proposed, with full context, before it executes. No irreversible action happens without explicit approval.
Kill switch. A single control stops all agent activity immediately. If you see something unexpected happening, you can halt everything in one step without navigating through settings.
Audit log. Every session generates a complete audit trail. You can review exactly what the agent did, in what order, using which data, and with what outputs. This is not just a debugging feature — it is the accountability mechanism that makes enterprise adoption plausible.
Sandboxed execution. Cloud-side tasks execute inside isolated environments with their own filesystems and browser contexts. Local operations on your Mac mini are scoped to permissions you explicitly grant during setup. The agent does not get blanket access to your machine; it operates within a defined permission surface.
Local-first data residency. Because Personal Computer runs on hardware you own, your files never have to leave your machine for the agent to access them. Cloud models see the outputs and summaries the agent chooses to send; your raw data stays local by default.
This architecture acknowledges a core tension in agentic AI: the more capable the agent, the more damage a mistake (or a bad actor) can cause. The confirm_action gate, kill switch, and audit log give users meaningful oversight without requiring them to babysit every operation.
Beyond the consumer Personal Computer product, Perplexity announced expanded enterprise capabilities for Perplexity Computer at Ask 2026.
Enterprise features include:
Enterprise pricing is separate from Personal Computer. Perplexity Enterprise starts at $40 per user per month ($400 annually), with Enterprise Max at $325 per seat per month ($3,250 annually). The Enterprise Max tier gives teams access to the full Computer agent stack with enterprise compliance features.
The enterprise push is significant because it signals that Perplexity is not just chasing individual power users with Personal Computer — it is building toward a comprehensive enterprise AI infrastructure play that competes directly with Microsoft Copilot, Google Duet AI, and Salesforce Agentforce.
The PYMNTS coverage noted that Perplexity Computer had already completed what one enterprise customer described as 3.25 years of work in four weeks — a figure that, if substantiated at scale, would make the ROI case for enterprise adoption nearly impossible to ignore.
Personal Computer sits inside the Perplexity Max subscription tier, priced at $200 per month.
For that price, Max subscribers get:
The $200/month price point is not cheap. It positions Personal Computer as a productivity tool for professionals and knowledge workers who can justify the cost through time savings. For someone billing $150-$300/hour, a single hour of reclaimed time per day covers the monthly fee.
The waitlist structure means Perplexity is doing a controlled rollout rather than opening the floodgates. This is prudent: always-on agents with write access to email and Slack at scale create customer support and reliability challenges that a phased rollout manages more gracefully.
To join the waitlist, you need an active Perplexity Max subscription. If you are already on Max, the option to apply for Personal Computer access should appear in your account settings.
Personal Computer does not exist in a vacuum. Several products compete for the "always-on local AI agent" space:
Anthropic Claude with computer use. Claude has had computer use capabilities in API form since late 2024, letting it control a web browser and desktop. But Claude computer use is session-based and requires external orchestration infrastructure. Personal Computer is a packaged, always-on product with pre-built integrations.
OpenClaw (OpenAI). OpenAI's autonomous agent product operates in the cloud, similar to original Perplexity Computer. It does not have a local hardware component. Personal Computer's local-first approach differentiates it for users whose workflows depend on local files and apps.
Rabbit r1 / Humane AI Pin. Both tried the dedicated AI hardware approach and struggled with limited capability and high prices. Personal Computer takes the opposite bet: use commodity hardware (Mac mini) that already exists and run powerful cloud AI through it.
Zapier/Make with AI. These workflow automation platforms added AI steps to their no-code automations. They are trigger-based and rule-following rather than objective-driven. Personal Computer's reasoning layer means it can handle edge cases and novel situations that a static automation rule would fail on.
The closest competitive analog is probably what Microsoft is building with Copilot integrated into Windows — a system-level AI that has access to your local environment and cloud services. But Microsoft's approach is tightly coupled to Windows and the Microsoft 365 stack. Perplexity's model-agnostic, Mac-based, multi-integration approach offers more flexibility for users who are not all-in on the Microsoft ecosystem.
Personal Computer is a meaningful product announcement, but it is also a thesis statement.
The thesis: the next computing paradigm is not apps you open, it is agents that run. You do not launch a tool when you need it — you set an objective, and a persistent agent pursues it across all your data, all your services, and all your time zones, continuously.
This has been discussed in AI research circles for years. Personal Computer is one of the first consumer-facing attempts to make it tangible and purchasable. If it works — if the agent reliably executes objectives, handles edge cases gracefully, and earns user trust — it will validate the paradigm and accelerate similar offerings from every major AI lab.
The "AI OS takes objectives" framing from Srinivas is the key signal. Perplexity is not describing itself as a search engine that got better. It is describing itself as operating system infrastructure for the AI era. That is a much larger market claim, and Personal Computer is the first hardware-adjacent product that makes the claim concrete.
The Mac mini angle is also worth watching. Apple has not announced anything comparable — no "AI agent mode," no persistent local AI infrastructure product. If Personal Computer succeeds at scale, Apple will face pressure to respond, either through native features in macOS or by deepening partnerships with AI labs.
The 24/7 nature of the product is the genuine innovation here. Most AI tools are reactive: you prompt, they respond. An agent that monitors your Slack, your inbox, and your GitHub continuously and acts on your behalf without being asked represents a qualitatively different relationship between humans and AI software.
What is Perplexity Personal Computer? It is software that runs continuously on a Mac mini, giving AI agents persistent local access to your files, apps, and cloud services (Gmail, Slack, GitHub, Notion, Salesforce) to execute tasks 24/7 on your behalf.
How is Personal Computer different from Perplexity Computer? Perplexity Computer is cloud-only — it never touches your local machine. Personal Computer runs on hardware you own and has direct access to local files and applications, not just data you explicitly upload.
What hardware do I need? Any Mac at launch, though a Mac mini is recommended for its low cost and always-on suitability. Windows support has not been announced.
How much does it cost? $200/month as part of the Perplexity Max subscription. This includes 10,000 monthly computational credits.
Which AI models does it use? Claude (Anthropic), Gemini (Google DeepMind), and Grok (xAI) are supported. Users can configure which models deploy for which tasks and run multiple models simultaneously.
Can the agent send emails on my behalf? Yes, but only after you confirm. The confirm_action safety protocol requires explicit approval before any sensitive action — sending emails, making purchases, or deleting files.
What if I want to stop the agent immediately? A built-in kill switch halts all agent activity instantly. You do not have to navigate through settings.
Does Perplexity see my local files? Local files stay on your Mac mini by default. Cloud AI models see summaries and outputs the agent chooses to share, not your raw files. You control the permission surface during setup.
Is there an enterprise version? Yes. Perplexity Computer for Enterprise includes SOC 2 Type II compliance, SAML SSO, audit logs, and isolated sandboxing. Enterprise pricing starts at $40/user/month, with Enterprise Max at $325/seat/month.
How do I join the waitlist? You need an active Perplexity Max subscription. The waitlist option appears in your account settings once Max is active.
When will it launch to everyone? No general availability date has been announced. Perplexity is doing a phased rollout through the waitlist.
What integrations does it support at launch? Gmail, Slack, GitHub, Notion, and Salesforce. More integrations are expected as the product exits the waitlist phase.
Can I use Claude and Gemini at the same time? Yes. You can configure the agent to deploy multiple models simultaneously, with different models handling different task types.
Does it work on a MacBook, or only a Mac mini? It works on any Mac. The Mac mini is recommended because it can run continuously without draining a laptop battery or occupying your primary machine.
What happens to the audit log? Every session generates a complete activity log that you can review. This serves both as a debugging tool and a compliance record.
Is this the same as Perplexity's "Hey" feature on Samsung phones? No. Hey Perplexity is a voice assistant feature for Samsung devices. Personal Computer is a local AI agent platform for Mac hardware.
Can it make purchases on my behalf? Purchasing is classified as a sensitive action and requires explicit confirmation before execution.
What is the confirm_action protocol? Before any irreversible or sensitive operation, the agent surfaces a clear description of the proposed action and waits for your approval. Nothing happens automatically.
Why a Mac mini and not a dedicated AI hardware device? Dedicated AI hardware devices like Rabbit r1 and Humane AI Pin struggled with limited capability and high prices. Perplexity's approach uses commodity hardware that already exists, keeping costs low while leveraging powerful cloud AI for heavy lifting.
Where can I learn more? Perplexity's official product pages and coverage from Axios, 9to5Mac, Digital Trends, Neowin, Apple Insider, and Macworld have detailed reporting on the product.
OpenAI launches Frontier, a new platform for coordinating teams of AI agents across complex enterprise workflows, targeting a market projected to reach $263B by 2035.
Facebook Marketplace now lets Meta AI respond to buyer messages and auto-generate listings from photos, rolling out to 900M+ monthly users.
Meta plans four new MTIA chips through 2027, aiming to cut AI compute costs by up to 30% and reduce reliance on Nvidia hardware.