The conference was held inside a former church in San Francisco's North Beach neighborhood. The venue felt appropriate. On March 11, Aravind Srinivas, CEO of Perplexity, stood before a room of developers and declared, "A traditional operating system takes instructions. An AI operating system takes objectives." Then he showed them what he meant.
The announcement: a product called Personal Computer — software that runs continuously on a user-supplied Mac mini, merging local files, applications, and active sessions with Perplexity's cloud AI engine. No special hardware required. Just a $599 Mac mini, always plugged in, always awake, always working.
The crowd in the church had questions. So did Wall Street.
The Distinction That Matters
Before Ask 2026, Perplexity already had a product called Computer — a $200-per-month cloud service that orchestrates 19 frontier AI models (Claude, GPT series, Gemini, and others) to complete multi-step research, coding, and business tasks. That version lives entirely in the cloud.
Personal Computer is different. It is software you install on a physical Mac mini you own and keep running on your desk, in a closet, or under a TV. The device acts as a persistent local proxy, giving Perplexity's AI agents continuous access to your local file system, running applications, and open browser sessions — things a purely cloud-based agent cannot reach without screen sharing or manual uploads.
Every action the agent takes on that machine requires your explicit approval first. Every session generates a full audit trail. There is a kill switch that shuts everything down immediately.
The Mac mini was chosen deliberately. It is cheap, nearly silent, low-power, and designed to run unattended. Apple's M4 chips mean it handles local context operations without breaking a sweat. Perplexity is not selling you a new computer. It is asking you to repurpose one you might already have sitting in a drawer.
Access is gated behind the Perplexity Max subscription tier, currently priced at $200 per month, and is rolling out through a limited waitlist as of mid-March.
What Computer for Enterprise Actually Does
Alongside Personal Computer, Perplexity launched Computer for Enterprise at Ask 2026 — a version built specifically for companies rather than individuals.
The enterprise product connects to more than 400 business applications, including Snowflake, Salesforce, HubSpot, SharePoint, GitHub, Linear, Gmail, Outlook, and Notion. Administrators can install additional connectors using the Model Context Protocol. Employees can query @computer directly inside Slack channels and threads, then continue those conversations in Perplexity's web interface or mobile app.
The compliance stack is full enterprise-grade: SOC 2 Type II certification, SAML single sign-on, isolated sandboxing per query, and full audit logs. Perplexity states that enterprise customer data is not used to train its models.
On the finance side, the platform can now pull live data from over 40 sources, including SEC filings, S&P Global, FactSet, Coinbase, LSEG, and Quartr.
Three new developer APIs were announced at the conference — Agent, Embeddings, and Sandbox — joining the Search API that launched in September 2025. Each handles a specific function: Search for grounded retrieval with cited outputs, Agent for multi-step task delegation, Sandbox for controlled execution, and Embeddings for semantic retrieval systems.
Perplexity cited an internal study covering more than 16,000 queries, claiming Computer for Enterprise saved teams the equivalent of 3.25 years of work in four weeks and roughly $1.6 million in labor costs.
The Startup That Replaced Its Entire Marketing Stack
The most viral story from Ask 2026 did not come from Perplexity's own demos. It came from a startup that took the stage to describe what it had done with an early build of Computer for Enterprise.
The company claims it replaced $225,000 worth of annual marketing tools in a single weekend. The agent scanned advertising dashboards hourly, managed budgets across platforms, detected campaign fatigue before it impacted spend, and made 224 micro-optimizations to the ad stack during one test run.
The claim has not been independently audited. Perplexity shared it as a case study. Whether it holds up at scale, across different industries, and with less technical teams remains to be seen. But it landed exactly the way Perplexity intended: as a direct threat to every SaaS vendor in the enterprise marketing stack.
Who Loses
The threat is structural, not just competitive. Enterprise software is priced per seat — you pay per person who uses the tool. If an AI agent can do the work of ten people, a company might need ten seats instead of a hundred. That is not a pricing negotiation. It is a category collapse.
Salesforce's stock has dropped 26% year-to-date. The phrase "SaaSpocalypse" has appeared in multiple analyst notes. The anxiety is not new — it predates Perplexity's announcements — but Ask 2026 gave the narrative a sharper edge.
Microsoft 365 Copilot requires an existing Microsoft 365 license, which runs $6 to $36 per user per month on top of the Copilot add-on. Salesforce Einstein is deeply embedded in Salesforce CRM contracts, making it both powerful and expensive to reach. Perplexity is entering this market at $200 per month per user for the full system — no legacy license required, no existing platform dependency.
| Feature | Perplexity Computer | Microsoft 365 Copilot | Salesforce Einstein |
|---|---|---|---|
| Price | $200/month (Max tier) | $30/user/month + M365 license | Bundled into Salesforce contracts |
| Deployment | Local (Mac mini) + Cloud | Cloud only | Cloud only |
| Local file access | Yes (via Personal Computer) | No | No |
| Audit trail | Yes (full session log) | Partial (admin logs) | Yes |
| Integrations | 400+ apps via connectors | Deep Microsoft 365 suite | Deep Salesforce CRM suite |
| Model selection | 20+ frontier models, auto-routed | Microsoft/OpenAI models | Salesforce-hosted models |
| Slack integration | Yes (@computer in channels) | Via Teams only | Via third-party connectors |
| SOC 2 Type II | Yes | Yes | Yes |
What Skeptics Say
Not everyone at Ask 2026 walked out convinced.
The most direct criticism: Personal Computer is not fundamentally different from a managed local agent you could build with existing tools. OpenClaw, NanoBot, and similar open-source agentic frameworks have been running local workflows for months. The question is whether Perplexity's polish, model routing, and connector ecosystem justify a $200-per-month subscription when open alternatives exist.
The more serious concern is trust. Enterprise IT departments are being asked to route sensitive Snowflake queries, legal documents, and proprietary business intelligence through a three-year-old startup's infrastructure. Perplexity's SOC 2 certification helps, but it does not close the gap with companies that have spent years building data governance relationships with Microsoft or Salesforce.
Amazon's legal pressure adds another layer of friction. On March 10, a federal judge granted Amazon a preliminary injunction blocking Perplexity's Comet browser agent from accessing password-protected Amazon accounts. The ruling turned on a distinction that will define agentic AI broadly: a user may authorize an agent to act on their behalf, but that is separate from whether the third-party platform authorizes the agent's access. Perplexity filed an appeal on March 11 — the same morning as Ask 2026. That timing was not lost on enterprise buyers evaluating how much legal exposure comes bundled with a Perplexity deployment.
Privacy-conscious users raise a different concern. Personal Computer places a persistent, always-on agent process on a local machine that phones home to Perplexity's servers. Even with an audit trail and a kill switch, the agent has continuous visibility into local files, app states, and active browser sessions. That is a meaningful surface area for data exposure, whether through misconfiguration, a security incident at Perplexity, or scope creep as the agent is given more permissions over time. Perplexity's policy states enterprise customer data is not used for model training — but policy must be trusted, not verified.
The Bottom Line
Personal Computer is a genuine product with a coherent thesis: your computer should have persistent objectives, not just wait for instructions. The Mac mini framing makes that thesis tangible and cheap. The enterprise version makes it immediately relevant for any team paying five figures per year for tools that each do one thing.
What remains unresolved is whether a three-year-old search startup can earn the kind of institutional trust that enterprise IT departments extend slowly, after audits, procurement cycles, and legal review. Microsoft and Salesforce have decades of that trust. Perplexity has a former church in North Beach, a very good launch event, and a $200-per-month bet that executives will move faster than their procurement teams.
Some will. The startup that replaced its marketing stack in a weekend is proof. The question is whether that becomes a trend or stays an anecdote.
For anyone paying close attention to where enterprise software dollars go next, this is worth watching. AI agents that combine local machine access with cloud orchestration are not a distant possibility. They are already running. The audit trail will tell you what they did.
Sources
- Perplexity launches Mac-based AI agent — Axios
- Perplexity's Personal Computer is a Mac mini running an AI OS — Macworld
- Perplexity's Personal Computer is a cloud-based AI agent running on Mac mini — 9to5Mac
- Perplexity takes its 'Computer' AI agent into the enterprise — VentureBeat
- Perplexity turns your Mac mini into a 24/7 AI agent — The Next Web
- Perplexity: Everything is Computer — The Register
- Perplexity's Computer for Enterprise Completed 3.25 Years of Work in Four Weeks — PYMNTS
- Perplexity Wants Its "Personal Computer" AI to Live on Your Desktop — Technology.org
- Perplexity CEO explains Computer — Fortune
- This Perplexity AI agent says it replaced $225K marketing tools stack in a weekend — Storyboard18
- Amazon wins court order to block Perplexity's AI shopping agent — CNBC
- The SaaSpocalypse of 2026: Why Salesforce's 26% Plunge Signals a Seismic Shift — FinancialContent
- New Perplexity APIs give developers access to agentic workflows — The New Stack
- Perplexity Event: Ask 2026 Developer Preview — AI CERTs