Perplexity launches “Computer” agent, orchestrating multiple AI bots for users in real

Perplexity’s “Computer” agent arrives as the first cloud‑only, multi‑model digital worker aimed at the premium‑price segment of the generative‑AI market. According to a briefing with Perplexity executives, the service is restricted to the $200‑per‑month Perplexity Max tier and runs entirely in the company’s data centers, sidestepping the local‑hardware security concerns that have dogged rival tools such as OpenClaw [TechCrunch]. The architecture is deliberately walled‑garden: users submit a high‑level request, and the backend orchestrates a fleet of 19 distinct AI models—including Anthropic’s Claude Opus 4.6 and Google’s Gemini—to execute sub‑tasks, synthesize results, and return a finished artifact such as a dashboard, web app, presentation, or animated GIF [Pcworld].

The core innovation is the “sub‑agent” pattern, where the primary Computer agent spawns specialized workers to handle discrete problem domains. Perplexity’s own documentation shows example workflows that pull statistical data, retrieve legal citations, and perform financial analysis before rendering the output as an interactive website or visualization [TechCrunch]. By delegating each step to a model best suited for the task, the system claims to reduce the “prompt‑chaining” friction that users typically encounter when stitching together multiple LLMs manually. The company positions this as a “general‑purpose digital worker that operates the same interfaces you do,” echoing the language used by chief business officer Dmi [Pcworld].

Security and reliability were highlighted as differentiators. Because the entire pipeline executes in the cloud, Perplexity can enforce uniform access controls, audit logs, and sandboxing for each sub‑agent, a contrast to OpenClaw’s client‑side execution model that has attracted scrutiny over data leakage [Pcworld]. However, the rollout was not without hiccups: a planned live demo at the press briefing was cancelled at the last minute after engineers discovered critical flaws in the orchestration layer, prompting the company to postpone hands‑on testing until the bugs are resolved [TechCrunch]. The incident underscores the technical complexity of coordinating dozens of models in real time, a challenge that has limited the commercial viability of similar agentic platforms to date.

From a market perspective, Perplexity is betting that enterprise and power‑user customers will pay a premium for a turnkey solution that eliminates the need to evaluate, license, and integrate multiple AI providers themselves. VentureBeat notes that the service “coordinates 19 models,” a scale that exceeds most competitor offerings and suggests a strategy to lock users into a single subscription rather than a fragmented stack [VentureBeat]. The pricing—double the cost of Perplexity’s standard Max plan—signals confidence that the productivity gains from automated, multi‑model workflows will justify the expense for businesses that need rapid prototyping of data‑driven assets.

Analysts have pointed out that the “digital‑worker” model aligns with a broader industry shift toward AI‑as‑a‑service platforms that abstract away model selection and infrastructure management [Ars Technica]. By presenting a unified API and UI, Perplexity hopes to capture a segment of the market that is currently experimenting with open‑source agent frameworks on GitHub but lacks the operational maturity for production use. If the company can resolve the early‑stage reliability issues and demonstrate consistent output quality across its heterogeneous model pool, Computer could become a reference implementation for cloud‑native AI orchestration—potentially setting a new benchmark for how enterprises consume generative AI at scale.