Perplexity has introduced Teammates, a set of personal AI agents designed to take ownership of long-running work with minimal human supervision. Rather than answering a single prompt, these agents operate continuously inside Perplexity Computer, the company’s cloud-based agentic assistant, and only surface when a decision is required. The feature is reserved for subscribers to Perplexity Max, the company’s most expensive tier.
- Persistent delegation: Each Teammate receives a role or task, works in the background around the clock and reports back when output is ready or input is needed.
- Multi-agent orchestration: Teammates can coordinate with one another, manage other Teammates and recruit additional agents as the workload expands.
- Premium pricing: Access costs $200 per month or $2,000 per year, initially on the web with iOS availability, while Android and native desktop apps are promised for later.
From answer engine to agent workforce
The launch marks a strategic pivot for Perplexity, which built its reputation on a search-driven answer engine. Teammates push the company toward software that holds a job for days or weeks rather than resolving a query in seconds. Perplexity describes the agents as built for “ongoing, large-scale work,” a deliberate contrast with the one-off assignments handled by Computer.
Each Teammate runs in its own isolated environment, complete with separate browser sessions and encrypted saved credentials. Users can watch progress through dedicated threads in the sidebar of Computer, hand off additional work, redirect an agent mid-task or request a status update. The agents draw on the Projects, Memory and Skills a user has already built inside Perplexity and Computer, along with existing connectors, files and past sessions.
“Teammates build on your Projects, Memory, and Skills that already exist in Perplexity and Computer.” — Aravind Srinivas, Perplexity CEO
Integration goes beyond the Perplexity interface. Teammates connect to plugins including Slack, GitHub and Gmail, with per-repository GitHub permissions. Added to a Slack channel, an agent can pick up work from colleagues and react when something changes in the conversation. Perplexity’s own examples span marketing, sales, customer support and operations: an operations agent chasing missing items 72 hours before a deadline, a support agent reviewing the previous day’s complaints, or a legal research assistant reading a contract, comparing terms and drafting a response.
Autonomy comes with open questions
The ambition is considerable, and so is the risk. A team of agents that organizes itself and effectively hires new members is an appealing productivity vision, but the more agents operate without oversight, the harder it becomes to know what they are doing and which data they are touching. Perplexity has addressed part of that concern with isolated environments and scoped repository access, yet the review burden on the human operator remains substantial.
There is also the question of reach. At launch, the agents operate within Perplexity’s ecosystem and its supported connectors; broader access to third-party applications is not part of the initial rollout. A team of agents is only as useful as the tools it can actually reach, and Perplexity will need to prove that its assistants connect to the places where real work lives.
Pricing reinforces the positioning. At $200 a month, Teammates targets professionals for whom delegation carries a clear return, not casual users. Perplexity has framed the current access as a promotional period, with an undisclosed number of credits eventually required to run the technology. The company also says it will expand Teammates’ access and features in the coming weeks.
A crowded and costly race
Perplexity is not alone in pursuing always-on agents. OpenAI, Microsoft, Anthropic, Meta and SpaceX have all introduced comparable products, though several target organizations rather than individuals. Perplexity’s decision to sell Teammates to individual Max subscribers sets it apart, and its price sits above cheaper alternatives whose usage limits tend to arrive quickly.
Srinivas has described Teammates as a first step toward agents that remember users and their organizations, learn to work alongside colleagues and improve each time they are corrected. That is a meaningful shift in how AI products are sold: from sessions and prompts to standing responsibilities. Whether the underlying economics hold, and whether users trust autonomous agents with sustained access to their tools, will determine if Teammates becomes a genuine productivity layer or an impressive demonstration of where the category is heading.