Products & Tools

Perplexity Adds Voice Control for AI Teammates on iOS and Android

Perplexity's November 18, 2026 release lets Max subscribers assign, steer and review persistent AI teammates by voice on iOS and Android, with permissions and review gates kept manual.

Perplexity Brings Voice-Controlled AI Teammates to Mobile
Perplexity Brings Voice-Controlled AI Teammates to Mobilestevepb / pixabay
By James Calloway4 min read

Why it matters

  • Perplexity added voice mode for AI teammates on iOS and Android on November 18, 2026.
  • Nearly a third of the most demanding work on Perplexity Computer already starts on an iPhone, per Perplexity.
  • Teammates launched on the web on November 4, 2026 and retain goals, task history and shared files across runs.
  • Web and mobile teammates require a Perplexity Max subscription; Free and Pro users must upgrade.
  • Slack, GitHub and Gmail are among available plugins, with GitHub access limited to chosen repositories.

Perplexity on November 18, 2026 shipped voice mode for its AI teammates on iOS and Android, turning the phone into a remote control for agents that keep working after the call ends. The feature is limited to Max subscribers (Perplexity's top paid tier) and sits on top of the agent system the company launched on the web on November 4, 2026.

The release matters because it extends a specific bet: that AI agents become long-running workers needing supervision rather than one-shot chat queries. Perplexity supports that bet with a usage figure — nearly a third of the most demanding work on Perplexity Computer (the company's agent-driven desktop environment) already starts on an iPhone. Voice mode now covers both starting those assignments and steering them after the initial request.

What does voice mode actually do?

A user taps a call button to start an assignment, then can call back to discuss revised priorities. The teammate plans the work, checks its findings and returns results for review. It can also operate inside connected applications and send a notification when it needs a decision.

The user then has two options: answer aloud, or open the thread to inspect completed and upcoming work. Perplexity's framing emphasizes follow-through — agents that continue a project, report back and wait for direction rather than terminating after one response.

The promise is a phone as supervision layer. The caveat: every claim about reliability rests on Perplexity's own description, and the announcement offers no benchmark, error rate or third-party evaluation of how well the agents perform unsupervised work.

Which tasks does Perplexity say teammates handle?

The company's role examples stress assignments with monitoring and follow-up built in:

  • A research analyst teammate can examine market changes, identify new developments and update a plan daily.
  • A data analyst teammate can track signup-to-purchase conversion, locate where users drop off, recommend a test and report whether it helped.
  • A software engineering teammate can investigate and fix a checkout bug, add a test and show a preview before code is merged — then check whether the fix holds after release.

In each case, findings feed into the next action while the user reviews the outcome and sets direction. The software-engineering example is the most explicit about control: it places a human review before merging and keeps verification part of the assignment after deployment. That review gate is a design choice, not a guarantee — Perplexity describes the sequence, but does not say what happens if a user skips the review.

How are agents configured and secured?

Creating a teammate in the mobile app involves five steps: naming it, connecting tools, describing the work it should own, assigning it to a project and providing an initial task. The description can specify where work comes from, what the agent should deliver and when it should act.

Slack, GitHub and Gmail are among the available plugins. GitHub access can be limited to chosen repositories. Perplexity stresses that voice control therefore sits atop configured permissions — it does not replace decisions about which repositories, inboxes or other tools each agent can reach. Repository and inbox access remain deliberate, manual choices.

Web and mobile teammates both require a Perplexity Max subscription. Free and Pro subscribers can upgrade through their account's subscription settings. Perplexity did not disclose pricing changes or subscriber numbers alongside the announcement.

What is the agent system underneath?

The mobile feature builds on the web launch of teammates on November 4, 2026. These agents coordinate Computer sessions and other teammates while retaining project goals, task history and shared files across runs. They can monitor work, act without a fresh query and send progress updates or approval requests.

Each teammate runs in a separate environment with its own browser sessions and inherits Computer's existing tool and Project permissions. Saved credentials are encrypted, according to Perplexity. Users can inspect completed work, take over the desktop or redirect the agent at any point.

The architecture answers the obvious worry about persistent agents — unbounded access — with isolation and inherited permissions rather than new privileges. What Perplexity does not detail is how failures inside a delegated app action are caught, beyond the agent's own self-checking described in the workflow examples.

Why this lands now

The competitive context is a shift from chatbots to agents that hold state, run on schedules and operate across applications. Perplexity's design keeps three practical boundaries around each agent: subscription access, configured permissions and explicit review points before consequential actions like code merges.

Perplexity said it would expand teammates' access and features in the weeks following the announcement. The stakes are straightforward: if roughly a third of heavy Computer work already begins on an iPhone, the phone becomes the primary supervisory surface for agent labor — and the accuracy of those agents' unsupervised work, which the announcement leaves unmeasured, will decide whether that surface earns trust.

Source: unite.ai

Share this article:

More from James Calloway

James Calloway

Show full bio

News editor covering people, launches and day-to-day practice in artificial intelligence and machine learning at AI In Context.

749 articles

Related articles

  1. Perplexity launches Teammates: AI agents that own ongoing work, not just chats
  2. Perplexity Launches Multiagent Mode to Coordinate Frontier AI Models
  3. Twilio Adds Voice AI Tools Built for Live Customer Calls, Not Demos
  4. Ex-Plaud exec raises $26M for Ambi, an AI that listens all day
  5. Manus 2.0 turns the AI agent into a platform

« Previous articleNext article »