Perplexity has launched Perplexity Teammates, a set of always-on AI agents that take ownership of ongoing work instead of answering one prompt at a time. Announced in a company blog post and now rolling out as a research preview, Teammates are available to individual Perplexity Max subscribers on the web, with an iOS app promised next.
The pitch is simple: give an agent a job, such as watching a sales pipeline or triaging bug reports, and let it keep working in the background, checking in only when it needs a decision.
Key takeaways
- Teammates are persistent agents built on Perplexity Computer, the multi-agent system Perplexity launched in February.
- Each Teammate runs on its own isolated computer, with a separate desktop, browser and cookies.
- They can join Slack channels and connect to tools such as GitHub, Linear, Gmail and Datadog.
- New Teammates use Claude Opus 5.5 by default, but you can pick a different model per Teammate.
- Access is limited to individual Max subscribers ($200 a month) on the web for now; usage draws on Max’s monthly credits.
What happened
Perplexity describes Teammates as “personal agents for long-running work.” Rather than starting fresh with every chat, a Teammate keeps one persistent thread that holds a project’s goals, task history, shared files and the feedback you gave it earlier. It plans its own tasks, checks results along the way and hands back finished work for review.
The product sits on top of Perplexity Computer, which already splits a large assignment into smaller tasks and routes them to subagents. Teammates add the missing layer: continuity. A Teammate can coordinate several Computer sessions at once, hand work to other Teammates and keep an eye on connected tools without waiting for a new prompt.
Perplexity’s own examples show the idea. A sales Teammate notices that a regular customer has stopped ordering, reviews their order and support history, and drafts an offer. An operations Teammate learns that an agency needs two weeks of lead time and builds that into future schedules. In engineering, one Teammate can find bugs on a website while another prepares fixes as pull requests.
Setup happens on the web. You select Teammates in the left sidebar, name the agent after its project, connect the tools it needs, and describe the work it owns: its sources, deliverables and timing. Perplexity’s help center guide suggests adding a schedule if the job repeats.
How Teammates are built
The most interesting design choice is isolation. Each Teammate gets its own computer with a separate desktop and browser, and browsing sessions and cookies are not shared between Teammates. Saved credentials are encrypted and entered securely, and the agent itself cannot read them. You can watch a Teammate’s desktop, take it over or redirect it in the middle of a task.
Teammates inherit the plugins, skills and permissions you have already set up in Perplexity, and use only those. For GitHub, Perplexity recommends granting access to just the repositories the agent needs. For Slack, you choose the workspace, which channels it monitors and where it posts updates. Colleagues can then assign tasks or ask for status directly in the channel.
On the model side, the help center says new Teammates default to Anthropic’s Claude Opus 5.5. You can choose another model per Teammate and send different tasks to different models, and any subagents a Teammate starts use its model unless you say otherwise.
Pricing and access
Teammates are in research preview for individual Max subscribers on the web. Pro, Enterprise Pro and Enterprise Max accounts are not included by default; enterprise users need to ask their admin. Perplexity says it will expand access and features “in the coming weeks.”
Max costs $200 a month or $2,000 a year. Perplexity does not list a separate Teammates price. Instead, agents draw on the monthly Computer credits that come with Max, and the help center warns that complex projects, parallel agents, large codebases and frequent automations burn through more credits. ITdaily and Data Studios report that Teammates usage is free for Max users during a launch promotion, but Perplexity has not said when that ends.
Why it matters
Teammates is Perplexity’s entry in what has quickly become the hottest product category of late 2026: the persistent work agent. OpenAI has Dots inside ChatGPT, assistants that track project context and come back when they need input. Anthropic previewed Claude Duet in October, a proactive agent that runs on its own virtual machine. Meta has been pushing Muse further into long-running tasks too.
The shift is from session-based prompting to delegating goals. That changes what matters in an AI product. Raw model quality still counts, but so do memory, permissions, scheduling, cost controls and the ability to inspect what an agent did. Perplexity’s choice to run each Teammate in its own sandboxed machine, and to make its desktop viewable, is a direct answer to the trust problem that has slowed agent adoption.
It also shows how model-agnostic these products are becoming. Perplexity does not train the frontier model behind Teammates; it defaults to Claude Opus 5.5 and lets users swap. The value it is selling is orchestration, tools and an isolated runtime, not the model itself. We saw the same pattern when Meta bought Conductor, an app built for running parallel AI coding agents.
What it means for developers and businesses
For engineering teams, the GitHub and Linear hooks are the obvious draw: an agent that watches an issue tracker, reproduces bugs and opens pull requests without being re-prompted. Datadog support hints at on-call style monitoring. Treat these like any new service account. Scope repository access tightly, decide which channels an agent may post in, and review its output before merging.
For small businesses, Perplexity’s examples lean on marketing, sales, support and operations work that is repetitive but needs judgment. The big unknown is cost. Because credits do the billing, an always-on agent that checks a channel every hour can cost very differently from one that runs a weekly report. Watch the Usage page closely once the promotion ends, and set a spending cap.
Persistent agents also behave in ways that are hard to predict over long runs, a point made vividly by the Emergence World experiment. Start with narrow, low-risk jobs before handing a Teammate anything customer-facing.
What to watch next
- Pricing after the promotion: how many credits a typical always-on Teammate consumes per month.
- Wider access: when Pro and Enterprise plans get Teammates by default.
- Mobile: Perplexity has promised iOS first; other platforms have not been dated.
- Limits: Perplexity has not published caps on the number of Teammates or parallel agents per account.
Perplexity calls Teammates “a new way of working with AI.” Whether that holds up will depend less on demos and more on whether these agents stay reliable, affordable and easy to supervise once they are running for weeks at a time.