What Is Perplexity Teammates? Persistent AI Agents That Take On Ongoing Work

On 5 November 2026, Perplexity launched Teammates: AI agents that hold a job, remember what they have done, and keep working when you are not looking. Instead of answering a question and forgetting it, a Teammate takes on an ongoing responsibility — covering the on-call rota, reviewing contracts, running a storefront — and carries it forward day after day. This guide explains what Teammates are, how they work, what they cost, and how they compare with the rival agent products from OpenAI and Anthropic.

What Perplexity Teammates are

A Teammate is a persistent, named AI agent that you create inside Perplexity and give a role. You describe the job in plain language, set its standing instructions, and tell it when to bring a person in. From then on it owns that work. It plans the tasks, does them, and keeps a single thread that never closes, where you can read what it is doing, what it has finished, and what it is waiting on.

Teammates are built on Perplexity Computer, the company’s agent platform, which launched in February 2026. Computer is the machinery underneath: it breaks a project into smaller tasks, picks models for each one, and drives software on the user’s behalf. Teammates add the part that makes an agent feel like a colleague rather than a tool — a durable role, a memory, and the habit of starting work on its own when something changes.

The launch was announced by Perplexity chief executive Aravind Srinivas on X at 17:02 UTC on 5 November 2026. Teammates are rolling out first on the web to Perplexity Max subscribers. Native apps for iOS, Android and desktop are coming soon, and the product is in a research preview, which means it is an early release that Perplexity is still shaping with user feedback.

How a Teammate differs from a normal chatbot session

A chatbot session starts empty. You ask something, you get an answer, and when the chat ends the context mostly ends with it. A Teammate starts full. It inherits everything already in your Computer workspace — your projects, your skills, your memory, your plugins and your tool permissions — so it begins work knowing your coding conventions, your contract terms, or your campaign rules. That shift, from answering prompts to owning outcomes, is what separates an AI agent from a chatbot.

The second difference is time. A chatbot waits for you. A Teammate watches. It can monitor the tools you connect — Slack, GitHub, Linear and others — and start work when something arrives: a bug report, a review comment, an alert at 2 a.m. It schedules follow-ups, keeps its role until you change it, and retains memory between conversations. If a task stretches over days, the agent picks up where it left off instead of needing the whole story retold.

The third difference is reach. A chatbot answers in text. A Teammate acts in software. Each one gets its own desktop computer and browser, with its own sessions and cookies, and can work with apps and websites — including sites that have no Perplexity connector — and test a website from start to finish. You can watch what it does, redirect it mid-task, or take over its desktop yourself.

How a Teammate works

Setting one up takes a few steps, all on the web. You need an individual Perplexity Max subscription with access to the research preview, and you need access to the accounts and tools the Teammate will use. Hover over Teammates in the sidebar, choose the create button, give the Teammate a name or role, and pick its appearance. In the advanced options you can add instructions and starting tasks, describe the job — including a schedule, if the work should repeat — and then finish the setup in the chat that opens. After that, the Teammate lives in the sidebar next to your projects and history.

Once it exists, a Teammate works through several mechanics that Perplexity describes in its help centre guide to Teammates:

Shared context. Teammates in the same project share context, files and instructions while keeping their own roles. They can message each other when new information arrives, and a Teammate can break a goal into tasks, run them in parallel across several agents, and add agents as the work grows.

Its own computer. Every Teammate runs in an isolated desktop and browser, with separate sessions and cookies. A marketing agent can stay signed in to one service while an engineering agent works in a code repository, and neither sees the other’s session. Because the agent uses the permissions already configured for your Computer tools and projects, checking those permissions before it starts is essential: whatever access you have given, it inherits.

Proactive monitoring. A Teammate watches the tools you connect and acts when conditions change. It can respond to review comments after opening a pull request, or notice an alert and start investigating without being asked.

Approvals and escalation. You set the rules for when the agent must involve a person. In Perplexity’s on-call example, the agent takes the night shift, pulls logs when an alert fires, finds the cause, and opens a pull request with a fix and a rollback plan — but pages its owner if the fix needs approval, or if it cannot isolate the cause within 15 minutes. You can also tag Teammates as apps in Slack, GitHub and Linear, so they can reach a human colleague directly in the tool where that person already works.

Model choice. You can choose which model each Teammate uses, and direct it to use different models for different tasks. Perplexity says Computer routes subtasks among 19 AI models; the Teammate is the role wrapped around that routing.

The example roles Perplexity suggests

Perplexity offers three suggested roles, and its product page walks through each with a worked example.

On-call Engineer. The agent takes the night shift, watches the monitors, and reacts to alerts. In the demo, an alert fires at 11:42 p.m.: errors from the billing service after a deploy. The agent pulls the logs, traces the cause to a configuration change that dropped the retry budget to zero, and opens a fix with a rollback plan attached — asking for approval before anything ships. It then offers to take the Friday deploy as well.

Legal Research Assistant. Given a contract extension to review against the company’s standard terms, the agent reads the document, compares it clause by clause, flags anything non-standard, and drafts a redline with suggested changes for a person to review.

Storefront Manager. The agent runs the day-to-day work of an online shop: tracking orders, watching stock, and handling the routine tasks that keep a storefront running.

The product page adds two more patterns. An Ads Manager runs holiday campaigns inside spending caps, pauses ad sets that underperform, and sends anything needing a decision to a human. And an Engineer agent fixes a checkout failure, opens a pull request, and hands it to a QA Teammate, which tests the fix on desktop and mobile and reports back. The hand-off between agents — one fixing, another testing — is the clearest picture of what Perplexity is selling: not a smarter answer box, but a small team of specialists that pass work between themselves.

Perplexity’s chief executive says the company’s own staff have used about 10,000 teammates inside Perplexity over the past 30 days. That is a company claim about internal use, not an independent measure, but it signals that the product was tested on real internal work before launch.

Who can use Teammates, and what they cost

For now, the door is narrow. Teammates are available on the web to individual Perplexity Max subscribers. Max is Perplexity’s most expensive plan, at $200 a month, or $167 a month when billed annually. Enterprise Pro and Enterprise Max plans are not included by default in this launch, so companies on those plans cannot simply switch Teammates on.

On cost, there are two layers. A Max subscription includes 10,000 credits a month, and Teammates are free to use during a promotional period. After that, their work draws on the subscriber’s monthly credit allowance. Perplexity has not said how large the promotional grant is, or when it expires. That matters, because the real cost of a Teammate depends on how much it does: continuous monitoring, long browser sessions, retries after failures, and conversations between agents all consume credits. Until the promotion ends and real bills arrive, nobody outside Perplexity knows what a busy Teammate costs per month.

There is also the question of what a Teammate cannot yet do. There is no phone app yet — iOS and Android are promised, not shipped — so the agent lives on the web for now. And because Teammates inherit your existing tool permissions, setting them up well means auditing what they can reach: your code, your documents, your customer data, your production systems. The help centre is explicit that the agent operates inside the permissions you have already granted.

Teammates vs OpenAI Dots vs Anthropic Claude Duet

Teammates arrive in a crowded week for always-on agents. OpenAI launched Dots, its own always-on agents, at its DevDay event on 29 September 2026 — barely five weeks before Teammates. A Dot is an agent inside ChatGPT that runs on GPT-6 Astra, gets its own cloud computer and browser, keeps working toward a goal after the chat window closes, and connects to more than 4,000 apps through plugins. Dots roll out to ChatGPT Pro subscribers (from $100 a month, with the first dot included at no extra cost) and Business Premium, with an Enterprise beta that a workspace administrator must switch on. Anthropic, meanwhile, offers its own rival agent product in Claude Duet. The three products share the same pitch — an agent that keeps working — and differ in the details that decide whether a team can actually adopt one: price, which plan includes it, and how tightly it is woven into the tools a team already uses.

On price, OpenAI’s entry point is lower: $100 a month for ChatGPT Pro against $200 for Perplexity Max. On reach, both give each agent its own computer and browser, and both insist on approval gates for consequential actions. On home turf, each has an edge: Dots live inside ChatGPT, Slack and Microsoft Teams, where many teams already work, while Teammates inherit the research depth of Perplexity Computer — its projects, memory and skills — which suits work that starts from research rather than from chat.

The comparison that matters most is not between the products but between their promises and their proof. Tencent’s HY4, a computer-use agent model launched recently, shows the same architecture spreading beyond the US labs — every major lab now wants an agent that can operate a computer, not just talk about one. But agents that run for hours, hold credentials and act in production systems are a new kind of risk. Their makers will be judged on reliability: whether they keep accurate context over days, recover from failed tools instead of compounding an early mistake, and leave an audit trail clear enough to reconstruct what they did and why.

Limits and open questions

Teammates are a research preview, and several questions have no answer yet. First, cost: the size and expiry of the promotional grant are unpublished, so the monthly price of a working Teammate is unknown. Second, reliability: Perplexity’s demos show clean successes — a fault found, a contract redlined — but long-running agents in real workplaces will face messy tools, stale logins and ambiguous instructions, and there is no public record yet of how often they fail or need rescue. Third, security: an agent with broad inherited permissions is only as safe as the permissions it was given, and teams will need to decide how much access a Teammate should hold, which credentials it may touch, and how its actions are logged. The security questions around agent architecture — permissions, audit logs and failure recovery — apply directly here.

There are practical limits too. Mobile apps are still to come, enterprise plans are excluded by default, and the product is new enough that its rough edges — confusing escalations, wasted credits, noisy alerts — will only show up as more people use it. None of that makes Teammates a bad idea. It makes it an early one, best tried on bounded, low-risk work first: a recurring report, a monitored inbox, a draft that a person reviews before anything goes out.

What to watch next

Four things will decide whether Teammates matter beyond the launch headlines. The arrival of the iOS, Android and desktop apps, which would put a persistent agent in everyone’s pocket. The end of the promotional period, which will reveal the true running cost. Whether Perplexity opens Teammates to enterprise plans, which is where the largest teams — and the largest budgets — sit. And the first independent reports from users running Teammates on real work, outside Perplexity’s demos. If those reports show agents that finish jobs, escalate sensibly and cost a predictable amount, the chatbot-to-colleague shift will be hard to dismiss. If they show agents that need constant rescue, the always-on agent may stay a demo product a while longer.

For now, the shape of the change is clear. The question Perplexity is asking its users is no longer “what do you want to know?” It is “what job do you want done?” — and it is offering an agent that will keep doing it until told otherwise.

Sources: Perplexity — Teammates product page; Perplexity Help Centre — Get started with Perplexity teammates. Launch details cross-checked against independent reporting published 5–6 November 2026.

Related news coverage: Perplexity Launches Teammates, Always-On AI Agents for Work