Perplexity Launches Always-On ‘Teammates’ Agents for $200-a-Month Max Users

Perplexity this week launched Teammates, always-on AI agents that take ownership of ongoing work such as sales follow-ups, market research and operations. They plan tasks, run them in parallel and return results for human review. The company announced Perplexity Teammates on Nov. 4 for Max subscribers on the web, with iOS “soon.” The Verge noted that the Max plan costs $200 a month.

The launch puts Perplexity in direct competition with the agent products OpenAI and Anthropic shipped this fall. The Verge compared it to OpenAI’s Dots agents and Anthropic’s Claude Duet, noting that Perplexity’s version is more expensive.

How Perplexity Teammates work

Teammates sit on top of Perplexity Computer, the company’s system for running agent sessions on a user’s device and in the cloud. Each Teammate is a persistent agent with its own thread. It remembers project goals, task history and shared files, coordinates Computer sessions and other Teammates, and can message coworkers in Slack. Perplexity says Teammates monitor work and “investigate and act without waiting for a fresh query.” In one company example, a sales Teammate notices that a regular customer has stopped buying, reviews the order and support history, and drafts an offer or flags issues, requesting approval before it reaches out.

CEO Aravind Srinivas framed the ambition on X: “Over time, we envision every org having hundreds of teammates working for each person,” as quoted by The Verge. Perplexity’s Nov. 6 changelog adds a related option for Mac. Computer can now run tasks fully on supported Apple-silicon Macs, working across local files and apps such as Notes and Mail without sending personal data to the cloud. That mode is for Pro, Max and Enterprise users on Mac app 26.42.0 or later, and Perplexity recommends 36 GB of unified memory.

By the numbers: cheaper research, better-grounded answers

Perplexity paired the launch with two developer updates that come with company-reported numbers:

• Agent API effort levels (Nov. 5): Every level now costs less per query, the company says. The “medium” level rose from 55% to 77% on the public BrowseComp and DSQA research benchmarks while costs fell “almost 90%.” The “xhigh” level held its 76.5% score on WANDR, Perplexity’s own benchmark, at about 25% of the cost of its August 2026 setup. A new “ultra” level scored 85.9% on WANDR at $21.71 per task.
• Decisions API web grounding (Nov. 4): A new "mode": "web_grounding" setting runs a quick search before the decision model answers. On 328 true-or-false claims, accuracy rose from 56.4% to 97.3%, and errors fell from 143 to 9. Median latency rose from 139 to 366 milliseconds.

The grounding result is a gain of 40.9 percentage points, and errors fell by about 94% (TBR calculations from Perplexity’s figures). All of these figures come from Perplexity’s own tests. WANDR is a Perplexity benchmark, and the company notes that its 51% baseline on the public PWS-Decisions test matches the benchmark authors’ own run.

How it compares

Perplexity is entering a crowded field. OpenAI’s Dots and Anthropic’s Claude Duet, both introduced this fall, also run multiple tasks at once on a user’s behalf, The Verge noted. Perplexity’s pitch rests on two things it has emphasized since launching its answer engine: accuracy grounded in web search, and the ability to switch among models from different companies inside Computer. The Teammates post says the agents are “built on Perplexity’s focus on accuracy and Computer’s multi-model capabilities,” and that they “plan the work, check results along the way, and bring completed work back for review.” The company’s developer releases this week point in the same direction. Each one ties lower cost or higher accuracy to more searching and checking, not to a single bigger model.

What to watch

Perplexity has not published usage caps for Teammates beyond the Max plan requirement, and the iOS date is unspecified. For businesses, the practical questions are the same ones raised by rival agent products: what an always-on agent may change without approval, and how its spending is controlled. Perplexity says Teammates run in isolated environments with access controls. For the broader race among agent makers, see our future of AI hub.

Sources

• Perplexity — Teammates are agents for complex, ongoing work, Nov. 4, 2026
• Perplexity — Improvements to every Agent API effort level, Nov. 5, 2026
• Perplexity — Web grounding for the Decisions API, Nov. 4, 2026
• Perplexity changelog — Teammates, fully local Computer on Mac, and Voice mode, Nov. 6, 2026
• The Verge — Perplexity launches its own take on AI agents for work, Nov. 5, 2026