Perplexity launches Teammates: AI agents that own ongoing work, not just chats

Perplexity has launched Teammates, an AI agent feature that takes ownership of ongoing work rather than answering one-off prompts, and it's available starting November 9, 2026, as a research preview for individual Max subscribers on the web.

The scope is deliberately narrow. Access requires an individual Perplexity Max subscription. Pro users are excluded, and Enterprise Pro and Enterprise Max are not included by default, according to the company.

The stakes here are bigger than one feature launch. The release underlines how AI companies are trying to move beyond chat-based assistance toward systems that can manage persistent assignments across multiple software tools while keeping a human review step before work is finalized. Perplexity is betting that the next phase of AI products looks less like a search box and more like a staff of software workers.

What can a Teammate actually do?

Users assign each teammate a role — monitoring bug reports, tracking metrics, or researching a market. The agent then plans tasks, carries them through, and returns the result for review. That last part matters. Nothing lands without a human looking at it first.

The product builds on Perplexity Computer, the company's existing system for tool use, files, and code. This version adds support for ongoing tasks, larger projects, and coordination across multiple agents.

The multi-agent mechanics are the interesting part. When a user assigns a broader goal, a teammate can break the work into separate tasks, identify which can run in parallel, and start additional agents to handle them. Each agent operates on its own computer and carries out its assigned task from start to finish before the results are collected and checked.

A teammate also doesn't stop when the assignment ends. It can continue monitoring a project after the work is completed and make updates based on what it finds. Output can include reports, pull requests, or configuration changes — all subject to review.

Which tools do the agents connect to?

Teammates can connect to services including Slack, GitHub, Linear, and Gmail through Perplexity's plugin system. The workflow is conversational: a user asks a teammate to connect a service from within chat, approves the requested permissions, and then instructs it to monitor channels, repositories, or workflows.

The agents aren't limited to services with dedicated connectors. Teammates can work with apps and websites through their own desktop computer and browser, including sites that don't offer an integration. When a human needs to intervene, users can take over and steer the work directly.

Within projects, teammates share context, files, and instructions while keeping separate roles. They can also message each other when new information arrives — a sign, Perplexity suggests, of a more collaborative model of agent-based work within a single task environment.

How much does it cost to run?

The service is tied to Perplexity's subscription and credit system. Max subscriptions include a monthly credit allocation that teammates can draw on, and the company is explicit that heavier use costs more. Complex work may consume more credits, especially when:

• several agents run in parallel
• agents process large files or codebases
• automations run frequently

Users keep control over the underlying models. They can choose which model a teammate uses and can direct it to use different models for different tasks. New teammates default to Claude Opus 5.5, and any agents started by that teammate use the same model unless the user specifies otherwise.

What are the security limits?

Teammates inherit the permissions of installed plugins and project settings. Perplexity says saved credentials are encrypted and entered securely, and each teammate runs on an isolated computer with its own desktop, browser session, and local cookies — meaning agents don't share a browsing identity with each other or with the user.

Project permissions continue to apply when teammates work inside a shared project. A teammate's chat is visible to its owner, while shared chats are read-only for others. That said, other people can still work with a teammate in the Slack channels it monitors, which opens a path for team-level use even in an individually scoped preview.

What use cases does Perplexity intend?

The company's documentation sets out several intended applications:

• competitive research
• issue investigation across GitHub, Linear, and Datadog
• software optimisation work that tests several approaches in parallel and reports measured results

That last use case is the clearest signal of where this product wants to go. Testing several approaches in parallel and reporting measured results is work that traditionally required an engineer's time; Perplexity is positioning agents as a way to parallelize it.

Why this launch matters

The AI industry's center of gravity has been shifting from chatbots to agents for the past two years, and Teammates is Perplexity's most concrete step in that direction. The feature combines three trends in one product: persistent agents that keep working after a task ends, multi-agent coordination where one teammate can spawn others, and deep tool access across Slack, GitHub, Linear, and Gmail with a human review checkpoint before anything is finalized.

The restrictions tell their own story. By limiting the preview to individual Max subscribers — excluding Pro and enterprise tiers by default — Perplexity is testing the technology with its most committed users before rolling it into team environments where the stakes, and the security surface, are larger.

The credit-based usage model is also a quiet admission that agent work is expensive. Running multiple models in parallel, processing codebases, and maintaining ongoing monitoring all consume compute, and Perplexity is passing that cost structure directly to subscribers through its allocation system.

Whether Teammates graduates from research preview to a full product, and whether it expands beyond individual Max plans to the enterprise tiers it currently excludes, will say a lot about whether persistent, multi-agent work is a real product category or a promising demo.