On November 5, 2026, Perplexity announced Teammates: Max subscribers can create named, persistent AI agents with fixed responsibilities to look after recurring work. Unlike a one-off chat, each Teammate has its own memory, tools, and escalation rules, runs tasks in the background, and reaches out over Slack when a decision needs a human. The feature launches on the web first, with iOS and Android to follow.
Each Teammate gets its own desktop, browser session, and cookies, signed in separately so they do not interfere: a marketing Teammate can stay logged into Meta's ad platform while an engineering one watches GitHub. They inherit the user's existing Computer projects, skills, plugins, memory, and permissions, starting out with the team's coding conventions, brand requirements, or contract terms already in place, instead of being briefed from scratch every time. Teammates can also hand work to each other: in the demo, an engineering agent fixes a password-reset crash, passes the pull request to a QA agent for desktop and mobile testing, and takes the results back for review; in the on-call demo, an agent spots a spike in HTTP 502 errors after a deployment, traces it to a dropped retry budget, and drafts a pull request with a rollback plan.
The dividing lines are clear: access requires the $200-a-month Max plan, which includes 10,000 credits a month, and the launch adds an extra Teammates usage grant whose size and expiry have not been disclosed. How fast sustained monitoring, browser sessions, and multi-agent work burn through credits will only become clear once that grant runs out. Permission inheritance cuts both ways: it saves setup, but a misconfigured permission hands the keys to a program that works on its own. Perplexity's advice is to start with tightly bounded workflows and keep human approval for consequential actions.
Max users can create their first Teammate on the web now, assigning it a role, standing instructions, tools, and escalation thresholds. The 15-minute rule from the demo is worth copying: if the agent cannot isolate a cause within 15 minutes, it escalates to a human. Write down when a person must be involved before letting it run.
