What are Perplexity Teammates?
Perplexity announced Teammates in a company blog post on 4 November 2026, and The Verge reported the rollout the following day. The product is a set of named AI agents that run inside Perplexity Computer, the company's cloud-based agent system. Unlike a chat session, a Teammate is meant to own a job over time.
Perplexity describes the agents as "personal agents that can take on that long-running work." They "plan and coordinate projects, switch gears as priorities change, and present results for review." Users can inspect the work as it happens and redirect the agent before it finishes.
The core difference is persistence. Each Teammate remembers project goals, task history and shared files across runs, according to Perplexity. Feedback carries forward too. If a user tells an operations Teammate that an agency needs two weeks' notice, the company says it will build that lead time into later schedules.
Teammates also coordinate other agents. One Teammate can coordinate several Computer sessions and hand findings to another Teammate. Perplexity's example is an agent that finds bugs on a website and passes them to a second agent that prepares fixes.
Setup is deliberately concrete. A user names the Teammate, connects plugins such as Slack, GitHub or Gmail, and chooses which GitHub repositories it can reach. Then the user describes the work, assigns the Teammate to a project and gives it a first task. Perplexity says the agent can use existing Projects, skills and memory from Computer.
The launch is narrow. Teammates are available on the web for Perplexity Max subscribers, with an iOS version promised soon. Each Teammate gets its own thread in the Perplexity app, and users can also work with it in Slack. Free and Pro users see an upgrade path rather than the feature.
It helps to place the launch in Perplexity's product history. The company says it shipped the first web-capable AI assistant in Comet in July 2025, then built Computer for heavier work. Teammates are the next layer: not a new model, but orchestration that turns an agent into an ongoing assignment.
How much does Perplexity Teammates cost?
Teammates are not sold separately. The only way to use them at launch is a Perplexity Max subscription, which costs $200 a month on the monthly plan. Perplexity's pricing page lists Max at $167 a month when billed annually, which works out to about $2,004 a year. Paying monthly costs $2,400 a year.
Max is the top consumer tier. It includes "maximum Computer usage," 10,000 monthly Computer credits, 35,000 bonus credits, frontier models and priority access to new features, according to Perplexity. Pro costs $17 a month billed annually and includes 4,000 bonus credits, but it does not include Teammates.
That price gap is the first thing to understand. A team of five would pay $12,000 a year for monthly Max seats before buying any extra capacity. Perplexity does not publish a per-Teammate price or a team plan for the feature. The company directs Free and Pro users to Settings, then Account, then Subscription.
Credits matter because Teammates can run many tasks in parallel. Perplexity's help centre says Max plans start with 10,000 credits a month. ShiftDelete reports that running several agents at once drains those credits quickly. The practical cost of a Teammate therefore depends on how much work it does, not just the seat price.
For comparison, OpenAI includes dots on its $100-a-month Pro tier, and Anthropic includes Claude Duet in existing Pro, Max and Ultra plans. Perplexity's agent offering sits at the highest entry price of the three. The Verge put it plainly: Perplexity's take "is much more expensive."
The pricing also reveals who Perplexity is targeting. This is not a $20 add-on for casual users. Max is marketed for heavy, sustained work, with higher limits on research, asset generation and file uploads. Teammates inherit that positioning, so the buyers are likely small teams and power users rather than the mass market.
How do Teammates differ from Perplexity Computer?
Computer is the engine. Perplexity's help centre calls it an "autonomous AI worker" that runs in a "secure cloud/isolated sandbox" with persistent memory. It can research, use tools and produce documents, apps or other outputs. It is available on web, iOS and Android.
Teammates sit above Computer. Perplexity says they "build on the power of Computer" and work inside it with a user's tools and files. A Teammate coordinates Computer sessions rather than replacing them. It decides what to run next, checks results and keeps a project moving.
The distinction is ownership. Computer can handle a complex project when a user sets it up. A Teammate is configured to own an ongoing area, such as sales follow-up or marketing campaigns, and to act when conditions change. That includes investigating a drop in customer activity without waiting for a new prompt.
Slack is central to the design. Perplexity launched Computer in Slack on 1 April 2026, and the feature grew out of an internal tool that employees used in public channels. In its first four weeks, Perplexity says, the internal version performed "$1.6M of work for 300 Perplexity employees."
That figure is a company-reported internal metric, not an audited productivity study. Perplexity does not define how it valued the work. It is still a useful signal of how the company expects agents to be used: in shared channels, with colleagues adding context and reviewing output.
With Teammates, a user can add an agent to a Slack channel so colleagues can hand it work or ask for updates. Teammates can also pass tasks to each other. Perplexity says this shared direction is what turns an individual assistant into something closer to a colleague.
There is a practical consequence for managers. A Computer task usually has a clear start and finish. A Teammate has a remit, which means someone has to define success, review exceptions and decide when the agent should stop. That management work does not disappear; it moves up a level.
What can Teammates do, and where are the limits?
Perplexity lists four broad job families: marketing, sales, customer support and operations. A marketing Teammate can manage social posts, email campaigns, ads and SEO, monitor competitors and draft fixes when results slip. A sales Teammate can review inquiries and past purchases, prepare outreach and quotes, and flag support issues before a representative makes contact.
A support Teammate can summarise the previous day's complaints, flag cases that need a personal reply from the CEO, draft responses and track each case to resolution. An operations Teammate can watch deadlines, checklists, vendors and paperwork, then chase missing items 72 hours before a deadline.
Perplexity says each Teammate runs in a separate environment with its own browser sessions. Saved credentials are encrypted, and users can review completed work, take over the desktop or redirect the agent. Those controls matter because a Teammate may hold access to Slack, GitHub and Gmail at the same time.
The limits are just as important. Teammates are new, and Perplexity says it will expand their access and features "in the coming weeks." The launch is web-only for Max subscribers. For Teammates, Perplexity has not published reliability benchmarks, error rates or a breakdown of which model handles each step.
The product also changes the cost of a mistake. A wrong answer in a chat window is easy to ignore. A wrong message sent to a customer, or a bad change pushed toward a repository, is visible to colleagues and clients. Perplexity describes review and approval requests, but not a universal approval gate for every action.
Teams should therefore treat Teammates as supervised staff, not autopilot. Scope plugins tightly, limit repository access and define which actions need a human sign-off. The official setup flow already supports repository selection, which is a sensible place to start.
There is also a data question. An agent that learns from past work needs context, and context often includes customer records, internal threads and campaign performance. Perplexity says credentials are encrypted and environments are isolated. Buyers should still confirm what is retained, who can inspect a Teammate's thread and how access is revoked when someone leaves.
How does Perplexity compare with OpenAI dots and Claude Duet?
Teammates arrive in a crowded field. OpenAI launched dots on 29 September 2026. They are always-on agents with a cloud computer, included on the $100-a-month Pro tier and on Business Premium, with Pro availability excluded in the EEA, UK and Switzerland. OpenAI is also rolling dots out to Enterprise plans in eligible markets.
Anthropic opened Claude Duet to a waitlist on 21 October 2026. Duet is a proactive agent for Pro, Max and Ultra subscribers in the US, included in their plans and subject to the same usage limits. It runs in a virtual machine with its own browser. Anthropic says an auto-mode classifier reviews actions, and some steps wait for a user's approval.
The three products share a template: persistent memory, a cloud desktop, proactive monitoring and some form of human review. The differences are packaging and price. Dots start at $100 a month. Duet adds no separate subscription for existing Claude subscribers. Teammates require the $200 Max plan.
Perplexity's bet is that collaboration is the differentiator. Teammates are designed to work in Slack channels, hand work to each other and follow business functions rather than single tasks. Perplexity also advertises more than 400 app connectors. That breadth is useful, but it also widens the surface area for mistakes.
The pricing gap may be defensible if a Teammate replaces a meaningful slice of coordination work. It is harder to justify for a small team that mainly wants a persistent assistant. Buyers should compare the total cost of credits, not just the headline seat price.
All three vendors are also learning in public. Each launch has been followed by updates to models, features and access. That pace is a reason to pilot rather than standardise, at least until the products and prices settle.
What should teams check before adopting always-on agents?
The obvious use case is work that repeats but rarely fits a fixed script. Pipeline reviews, campaign monitoring, complaint triage and vendor follow-up all involve reading context, deciding what changed and taking a next step. Those are the tasks Perplexity highlights.
The less obvious risk is management load. An agent that owns ongoing work needs goals, boundaries and review. If one person can supervise "hundreds of teammates," as Perplexity's chief executive suggests, the bottleneck becomes judgment rather than execution. Someone still has to decide which customer gets an offer and which bug gets fixed first.
That is the argument in Prediction Machines, which frames AI as a steep fall in the cost of prediction while judgment remains scarce. Teammates make prediction and coordination cheaper. They do not remove the need for someone to set the objective and accept the consequence.
Human + Machine makes a related point about redesigning workflows around human-agent collaboration rather than bolting agents onto old processes. A team that adds Teammates to an unchanged approval chain may get faster drafts and slower decisions. A team that redesigns the hand-off can remove work entirely.
The Second Machine Age provides the wider context: automation shifts the boundary of what people do, and the gains depend on how quickly organisations adapt. Perplexity's internal Slack experiment is a small example. The agent improved because colleagues could see better ways to use it and pull it into more shared workflows.
Before adopting, teams should run a narrow pilot. Pick one recurring workflow, connect the minimum tools, define an approval boundary and measure cost per completed task. Review the threads weekly. If the agent cannot explain what it did, the workflow is not ready for autonomy.
Procurement should ask a few direct questions. Which plan includes the feature, how many credits does a typical task consume, what happens when credits run out, and who owns the output when an agent acts? Perplexity answers the first question directly and publishes monthly credit allowances. The cost per task and ownership of agent output are less clear.
What happens next for Perplexity Teammates?
The next milestones are straightforward. Perplexity has promised an iOS version and broader access. The launch post does not mention team seats, admin controls or service-level guarantees for Teammates. Those details will decide whether the feature stays a power-user tool or becomes something an IT department can standardise.
Watch for reliability data. Perplexity's launch post is long on workflows and short on measured outcomes. Independent tests of task completion, error rates and cost per outcome would be more useful than another demo. The same is true for OpenAI and Anthropic, whose agent launches have similar gaps.
Watch the price. If Perplexity bundles Teammates into Pro or introduces a team tier, the calculus changes quickly. If credits remain the main constraint, heavy users may find that a $200 seat is only the starting point.
Finally, watch Slack. It is where Perplexity's agent strategy began, and it is where Teammates are meant to prove themselves. A shared channel makes an agent's work visible, which can build trust or destroy it faster than any benchmark.
For now, Teammates are a serious product with a narrow door. The idea is coherent: give an agent a role, memory and a place in the team. The open questions are whether the work holds up under independent scrutiny and whether enough teams will pay $200 a month to find out.