Perplexity has launched Teammates, a set of always-on AI agents that take on continuing work instead of answering one question at a time. Chief executive Aravind Srinivas announced the launch on X at 17:02 UTC on 5 November 2026, and the product is rolling out to Max subscribers on the web from today. Native apps for phones and desktop are coming later.
A Teammate is a named, persistent agent built for a standing job. Rather than writing a fresh prompt for each task, a user gives it a high-level goal — run the on-call rota, track a legal review, manage a storefront — and the agent plans the work, carries it out, and keeps the thread going between sessions. Each Teammate has a single thread that never closes, so its owner can read what it has done, what it finished, and when.
Teammates are built on Perplexity Computer, the company’s agent platform, launched in February 2026. They inherit what is already in a user’s Computer workspace: projects, skills, memory, plugins and tool permissions. That inheritance is the point. An agent that already knows a team’s coding style, contract terms or campaign rules can start work without being briefed from scratch each time — a contrast with the blank slate of a new chat session.
Each Teammate also gets its own desktop computer and browser, with independent sessions and cookies, so one agent can stay signed in to one service while another works somewhere else. They can watch connected tools such as Slack, GitHub and Linear, start work when something changes, and message human colleagues in Slack when a decision is needed. They ask for approval at defined points. In Perplexity’s on-call example, the agent watches the monitors overnight, traces a fault, drafts a code change with a way to undo it — and only wakes its owner when sign-off is required or the cause stays hidden after 15 minutes.
Perplexity suggests three starting roles: On-call Engineer, Legal Research Assistant and Storefront Manager. Its product page shows the pattern in action: an ads manager pausing underperforming campaigns inside a spending cap, an engineer reproducing a checkout failure and opening a fix, and a legal assistant returning a redlined contract. Teammates can also pass work to each other — one fixes a bug, another tests it — and Perplexity says its own staff have used about 10,000 teammates inside the company over the past 30 days.
Access is narrow for now. Teammates are in a research preview for individual Max subscribers, Perplexity’s most expensive plan at $200 monthly, falling to $167 a month on an annual bill. Enterprise plans are not included by default. Use is free during a promotional period, after which Teammates draw on the subscriber’s monthly allowance of 10,000 credits. Perplexity has not said how large the promotional grant is or when it ends, so the real running cost of a busy Teammate is still unknown. Details of setup, permissions and credits are set out in Perplexity’s help centre guide to Teammates.
The launch lands in the middle of a race between the big AI labs to turn chatbots into co-workers that keep working after the user closes the laptop. OpenAI introduced its own always-on agents, Dots, at its DevDay event on 29 September 2026, and Anthropic offers a rival agent product in Claude Duet. The contest now turns on something narrower than model quality: which agent can be trusted with real tools, real permissions and real money, day after day, without a person watching every step. Tencent’s HY4, another computer-use agent model launched recently, shows the same idea spreading beyond the US labs.
That trust is still unproven. Long-running agents can compound an early mistake, burn through credits on retries, or flood a team with alerts — and even the best AI systems still struggle with real work when judged on finished tasks rather than chat answers. What matters next is evidence from outside Perplexity’s own demos: how Teammates behave on phones, what they cost once the promotion ends, and whether companies — Anthropic among them — open them up to enterprise teams. For now, the shape of the product is clear: the chatbot is becoming a colleague with a job description, a memory and a shift that never ends.
