Manus 2.0: Giving Every Personal Agent Its Own Email, Phone Number, and Wallet

On September 28, 2026, Manus launched Manus 2.0 and a companion app called Cue. This is not a routine point release but a ground-up rebuild: Cascade is the new orchestration harness, Manus Studio adds editable video and games, Automations trigger tasks from events, Cloud Computer offers persistent compute, Remote Control lets you drive a desktop from your phone, and Cue pushes personal agents into new territory, with each agent getting its own email address, phone number, wallet, and dedicated computer. Manus says Cascade cuts token usage by 23.2%, task time by 28.2%, and cost by 32% versus the prior system.

From a chatty tool to an agent that runs your errands

From a chatty tool to an agent that runs your errands

Manus 2.0 is positioned as a shift from an agent you direct inside a chat box to an execution entity that runs on its own. The change is one of perspective: previously you typed a prompt, waited for output, and typed the next one; now you describe an objective, set boundaries, and the agent handles the routine. Around that shift, Manus 2.0 fills in several pieces at once: a smarter orchestration harness, an environment that runs persistently, automations woken by events, creative tools you can edit, and an identity for the agent itself in the form of email, phone, and wallet. The pitch is no longer how good this one answer was, but whether the agent can get things done unattended.

# The release at a glance. Facts are from Manus's September 28, 2026
# launch materials and follow-up coverage.

MANUS_2 = {
    "launched": "2026-09-28",
    "harness": "Cascade",
    "claimed_vs_prior": {
        "token_usage": "-23.2%",
        "task_time": "-28.2%",
        "cost": "-32%",
    },
    "components": [
        "Cascade (orchestration)",
        "Manus Studio (Video Editor, Game Dev)",
        "event-triggered Automations",
        "Cloud Computer (persistent compute)",
        "Remote Control / Computer Use",
        "Cue (personal-agent app)",
    ],
}

Cascade: the orchestration harness underneath

Cascade is the new backend and the technical core of the rebuild. Its pitch is direct: you describe the objective and define what success means, and the harness handles routine orchestration rather than making you hand-specify every tool call. Manus reports that, versus the prior system, token usage drops 23.2%, task time drops 28.2%, and cost drops 32%. That all three fall together is unsurprising: if the harness plans steps more accurately, calls tools more sparingly, and retries less, then spending fewer tokens and finishing faster are two sides of the same coin. For users, it means more tasks per budget and less waiting per task.

# Cascade is the new backend. The pitch is that you describe the
# objective, and the harness handles routine orchestration for you.

def run_task(objective, cascade):
    plan = cascade.plan(objective)          # decompose, don't micromanage
    while not plan.done:
        step = plan.next()
        result = cascade.execute(step)      # pick tools itself
        plan.update(result)
    return plan.output

# Manus says using Cascade instead of hand-specifying every tool call
# is what trims tokens, time, and cost together.

Cascade orchestrates, letting the harness handle the routine

Cue: personal agents with identity

Cue is the standalone companion app and the most interesting piece of the launch. It turns each personal agent into an entity with identity: its own email address, phone number, wallet, and a dedicated computer. With those, an agent can send messages on your behalf, make payments within a set budget, take phone calls, and collaborate with other agents in shared group chats to finish multi-step tasks. The point is that only when an agent has an identity external systems can recognize does it become able to deal with the outside world for you, registering for services, handling email, completing transactions. But being able to spend also means being able to spend while you are away, so budget guardrails must be first-class rather than an afterthought.

# Automations turn the agent from pull to push. Tasks can start
# from events, not only from a human prompt.

TRIGGERS = [
    "a new email arrives",
    "a Slack message is posted",
    "a calendar update lands",
    "a Notion page changes",
]

def on_event(event, agent):
    if event.matches(TRIGGERS):
        return agent.start(event.payload)   # fire the workflow

Automations and the Cloud Computer

Manus 2.0 moves the agent from pull to push. Automations let tasks start from events instead of a human prompt: a new email arriving, a Slack message being posted, a calendar update landing, a Notion page changing. The companion Cloud Computer provides persistent compute, a running environment that does not vanish when you close your laptop. Together they give an agent the ability to be always-on: waiting for events, waking, executing, and returning to wait, with no human in the loop. For automation, event-driven execution plus persistent runtime is the step from demo to production.

# Cloud Computer is persistent compute, sold as an add-on. Cue is
# the personal-agent app where identity starts to matter.

CUE_AGENT = {
    "email": "own address",
    "phone": "own number",
    "wallet": "budgeted payments",
    "computer": "dedicated",
    "collaboration": "shared group chats with other agents",
}

# Plans: $20 / $40 / $200 per month for 4,000 / 8,000 / 40,000 credits,
# plus 300 refresh credits a day. A Cloud Computer costs $30 or $50.

Event triggers: a new email or Slack message can start a task

Studio: editable video and games

Manus Studio targets creative work, with tools such as a Video Editor and Game Dev, and it emphasizes editability. This gets overlooked but matters: if an agent's output can only be accepted whole or redone from scratch, it struggles to enter professional workflows; if every segment of video or game element can be fine-tuned by a human, the agent upgrades from one-click generation to a collaboration tool. Manus's guidance is pragmatic: use Studio when the output needs iterative professional editing, and let Cascade handle the routine when the objective is clear and automation-first. Separating autonomous execution from human polish is the precondition for professional users to stay.

# The design question agents force on product teams: identity and
# spend limits must be first-class, because payments are irreversible.

def pay(agent, amount, memo):
    assert agent.wallet.budget_remaining() >= amount, "over budget"
    return agent.wallet.send(amount, memo=memo)

# Giving an agent its own wallet is powerful and sharp at once: it can
# act unattended, and it can also spend unattended. Budgets are the
# guardrail, not good intentions.

The bigger shift: agents as first-class economic actors

Put Cue's wallet, the Cloud Computer's persistence, and Automations' event triggers together, and Manus 2.0 points at a larger shift: agents are moving from software feature to economic actor. They have identity, accounts, budgets, and a persistent timeline, able to interact with outside services on your behalf. Pricing reflects that shape too: $20, $40, or $200 per month for 4,000, 8,000, or 40,000 credits, plus 300 refresh credits per day, with a Cloud Computer costing $30 or $50. That raises a question product teams cannot dodge: once an agent can pay and communicate on its own, security, authorization, and budget boundaries are not nice-to-haves but the foundation of whether the product can be trusted. And it raises a question for users too: how much of your digital life, from your inbox to your wallet, are you willing to hand to an agent that acts while you sleep? The answer will decide which of these products scale beyond early adopters.