Perplexity Teammates: always-on agents behind a $200 Max paywall

Perplexity has launched Teammates, persistent AI agents that take on ongoing work instead of waiting for the next prompt. They run inside Perplexity Computer and are available as a research preview to individual Max subscribers on the web. MLQ.ai, citing The Verge, reports that Max costs $200 a month on monthly billing. The price is less interesting than what it now buys: delegation of a responsibility rather than access to a model. The question of who answers for a bad call is still open.

What happened

Pulse 2.0 dates the announcement to 4 November 2026, while MLQ.ai gives 5 November 2026 as the launch of the research preview. Mobile access is coming later. Enterprise Pro and Enterprise Max accounts do not get the feature by default.

The setup is simple. You give a Teammate a role, connect the tools it needs and assign it to a Project. From there it plans, coordinates and presents work for review as priorities change. Launch documentation lists Slack, GitHub and Gmail among the integrations, and GitHub access can be scoped to individual repositories. A Teammate can monitor those tools, act when something changes and ask a person for input when it reaches an escalation point.

Teammates can also be added to Slack channels, where colleagues assign tasks and ask for updates. They can hand work to each other: Pulse 2.0’s example is one agent spotting defects on a website and another picking up the findings to prepare fixes. According to Dataconomy’s report on the launch, Teammates pull in existing projects, skills and memory from the Perplexity tools a user already has, so they don’t start cold.

On isolation, each Teammate gets its own desktop and browser environment. Credentials are saved encrypted and existing tool and Project permissions apply. Users can inspect completed work, redirect an agent mid-task or take over its desktop. Perplexity also says Computer tasks run in an ephemeral Firecracker microVM called SPACE. MLQ.ai is careful to describe these as company claims rather than an independent security certification.

On pricing, MLQ.ai notes that Max is $167 a month billed annually and includes 10,000 monthly Computer credits. Teammates draw on that allocation. Perplexity has published neither a separate Teammates price nor a credit cost for each type of agent activity.

What it means

This is a change of category. An assistant that answers questions is judged on the quality of each answer. An agent that owns a workflow is judged on outcomes over weeks: whether it chased the vendor, flagged the stalled deal, noticed the missing paperwork. That bar is much higher, and Perplexity is pricing for it.

The model question follows from that. None of the reports we reviewed name the model behind Teammates, and Pulse 2.0 describes Computer as a multi-model architecture. For a buyer, that is a problem. “Good enough for a search result” and “trusted enough to own recurring outcomes” are different standards. When a platform routes work across models, you can’t easily tell which model made a given decision on your live data. Every agent platform will face this question, and buyers should ask it directly.

Then there is accountability. MLQ.ai points to Forrester analyst Brian Hopkins telling ITPro in June that most enterprises were still unprepared to operationalise agentic AI, with orchestration, control, trust and auditability named as the main obstacles. MLQ.ai’s own conclusion is that Teammates are a meaningful expansion but not yet proof that persistent agents can run independently across sensitive systems. Perplexity has not published independent figures on reliability, task completion or error rates. When a Teammate gets a live decision wrong, it is unclear whether responsibility sits with the model, the platform or the person who delegated the work. The pricing shows that question is no longer hypothetical.

What it means for builders

The building blocks here are familiar to anyone shipping agents: proactive monitoring, explicit escalation points, handoffs between agents and persistent memory. Anthropic’s move toward agents that message you first points the same way. The hard engineering isn’t the agent loop. It is the approval design, which decides which actions run on their own and which wait for a human. Our notes on approval patterns that scale apply whether you buy a platform or build your own.

What to do now

If you have a Max seat and want to evaluate Teammates seriously, treat it like any pilot that touches production systems:

1. Pick one bounded, recurring workflow, such as the operations example Pulse 2.0 describes, where an agent spots outstanding requirements 72 hours before a deadline.
2. Scope permissions to the specific repositories, inboxes and channels the job needs, and nothing more.
3. Write the escalation points down before you switch it on, and name the human who owns the outcome.
4. Track credit consumption for a couple of weeks. Always-on monitoring has a running cost that Perplexity hasn’t itemised.
5. Log every wrong call and every override. That record is your only real reliability data until independent benchmarks exist.

Most pilots fail at the step from demo to dependable, and persistent agents make that step harder, not easier. Our piece on why AI pilots stall before production covers the usual causes. Decide what success looks like before the agent starts owning anything.

Frequently Asked Questions

How much do Perplexity Teammates cost?

Teammates require Perplexity Max. MLQ.ai, citing The Verge, reports Max at $200 per month on monthly billing, and Perplexity's pricing page lists $167 per month billed annually with 10,000 monthly Computer credits. Teammates draw on those credits, and no separate per-activity cost has been published.

Can Enterprise Pro or Enterprise Max users access Teammates?

Not by default. According to the launch documentation reported by MLQ.ai, Teammates launched as a research preview for individual Max subscribers on the web. Enterprise accounts need to request access or wait for a wider rollout.

Which tools can a Teammate connect to?

Launch documentation lists Slack, GitHub and Gmail among the supported integrations. GitHub access can be limited to specific repositories, and Teammates can be added to Slack channels to take tasks and give updates.