Perplexity has introduced Teammates, a feature that lets multiple AI agents coordinate with one another, resume where earlier sessions ended, and continue working without waiting for another prompt. The agents can also message coworkers through Slack. The central difficulty, the company says, is keeping everything synchronized when agents run at the same time, share project state, and operate with different permissions.
Announced on Wednesday, 11.11.2026, Teammates builds on infrastructure Perplexity has spent much of the year developing, putting that earlier work to use in workflows where agents delegate tasks and keep operating across sessions.
Keeping agents stateful
When Nate Kupp, Perplexity's vice president of infrastructure, showed The New Stack the company's agent sandboxes, he described how hard it is to maintain reliable systems at that scale, citing the durability and uptime demands of running millions of sandboxes.
"Stateful systems are incredibly hard to build," Kupp said.
Teammates relies on the existing infrastructure of Perplexity Computer, which keeps persistent project state separate from the individual environments where execution happens. That separation lets agents keep working on the same project without depending on one uninterrupted session.
In a technical breakdown of its shared workspaces, the company explained that it uses Git-backed repositories to preserve project state across sessions. Every Computer project gets its own repository, while agents operate in separate sandboxes and submit their changes back to the shared project.
Coordination becomes harder when several agents touch the same code at once: if one agent commits first, another may need to reapply its work on top of the newest version. Perplexity reports that under 2% of submissions are rejected for being based on outdated repository state. Median submission latency is under 1.7 seconds, and p99 remains below eight seconds.
Memory is managed separately through Perplexity's Brain system, which stores what agents have learned in dedicated repositories, so background agents can update shared knowledge without interfering with changes to project files.
Pausing and forking agents
Perplexity built SPACE around Firecracker microVMs and Btrfs snapshots, which makes it possible to pause, resume, or fork agent sessions without losing their saved state.
With Teammates, developers can connect tools such as Slack and GitHub and configure triggers that tell agents when to act. Perplexity has said little about what happens when that work stalls, particularly when one agent is waiting on another or acting on information that is already out of date.
Open permission questions
Each teammate runs inside its own isolated environment and browser session. Access is governed by Computer's existing permissions, and credentials are handled securely through encryption. Users can monitor progress and take control when needed.
What remains unclear is how the system stops agents from reaching resources beyond their permissions when tasks are handed off between them. For enterprises, that open question could shape how much autonomy they grant agents without requiring human approval.
Teammates is initially available on the web for Perplexity Max subscribers, with iOS support planned.