Manus AI in Singapore: what the Meta deal, and its reversal, mean for business

In this guide

• What is Manus AI?
• Who owns Manus AI now?
• Is Manus AI a Chinese company?
• How safe is Manus AI?
• Four checks for a Singapore business
• Can you use Manus AI for free?
• Personal agent subscription or deployed business system?
• When is Manus the right choice?
• How Olano helps

What is Manus AI?

Manus is an autonomous agent: you describe an outcome, and it plans the steps and carries them out on its own sandboxed cloud computer, browsing, running code and writing files until it hands back a finished report, spreadsheet, slide deck or website. Butterfly Effect, the team behind the earlier Monica browser assistant, unveiled it in early March 2025 as a "general AI agent" (5 March in China, which is why some listings say 6 March), and Manus calls the sandbox your cloud computer. On 28 September 2026 it released Manus 2.0, with event-triggered automations and a persistent cloud computer, plus Cue, an early-access app that gives each personal agent its own email address, phone number and wallet (Bloomberg, The Next Web).

What it is good at:

• Research that ends in a deliverable, such as a market scan written up as a report or a comparison table, rather than a chat answer you still have to turn into something.
• Building from a description: slides, spreadsheets, and simple websites and apps it can also publish.
• Long tasks you do not want to supervise. It works in the background and can repeat work on a schedule.
• Starting with nothing: a free plan, no setup, nothing to deploy.

Who owns Manus AI now?

Its founders and investors, not Meta. Butterfly Effect, led by chief executive Xiao Hong, said on 1 September 2026 that it had "formally resumed independent operations". Reuters reported on 8 October 2026 that the founders and earlier investors had helped buy it back from Meta at a reported US$2 billion valuation, and that it had closed a round of more than US$500 million co-led by Boyu Capital and IDG Capital, with Tencent, Sequoia China and ZhenFund taking part. Search results still lead with the acquisition, so here is the sequence.

Scroll the table sideways to see every column.

Where the sources disagree. Meta never disclosed a price, so "about" and "more than" US$2 billion are both reporting. Most coverage treats the purchase as closed in December 2025, hence an order to unwind rather than block. Some outlets name the investor Sequoia China and others HSG, the same firm under its newer name. A US$4 billion valuation was reported while Manus was raising; the company did not confirm the final figure.

Meta has since launched its own personal agent, Muse (8 September 2026, United States first), which we compare with a business deployment in Olano vs Meta Muse.

Is Manus AI a Chinese company?

It was founded in China and has been headquartered in Singapore since mid-2025, and it is available to users here. Its Singapore company, Butterfly Effect Pte. Ltd. (UEN 202330764R), appears on the ACRA register published through data.gov.sg as a live private company incorporated on 2 August 2023. Chinese regulators nonetheless treated the sale to Meta as within their review, which is why it was reversed. For a buyer the useful question is the paperwork rather than the flag: which entity you contract with, where your data is stored, and which law governs the terms, read in their current versions.

How safe is Manus AI?

Manus publishes real credentials: its security page lists SOC 2 Type II, ISO/IEC 27001 and ISO/IEC 27701 certification, and its trust centre names the United States and Singapore as data storage locations, with EU standard contractual clauses for international transfers. What certifications do not settle is continuity and authority. In August 2026 some users had to back up and restore their work because of the separation from Meta, a reminder that a change of owner can change what happens to your data. And no certificate answers who approved what an agent did.

Four checks for a Singapore business

They apply at the point where the work stops being one person's and becomes the company's.

1. Who owns the data now, and where it is processed

Read Manus's current privacy policy and terms, not coverage from the months it belonged to Meta. Under the PDPA, responsibility for personal data you put into any tool stays with your organisation, and customer lists, CVs and enquiry threads are all personal data. Where data is stored outside Singapore, as it is with a United States region, the PDPA expects it to be protected to a comparable standard. Ask for the data processing agreement and sub-processor list. Our PDPA guide for AI agents has eight questions worth putting to any vendor, including us.

2. Seats, and who is actually using it

The Team plan is priced per seat: from US$20 per seat per month or US$200 per year, with 4,000 credits per seat each month pooled across the team, guests who do not take a paid seat, and single sign-on free for teams of 30 or more (Manus help centre). That suits a defined group who each want their own agent. The trap is one personal login shared around the office: it loses any record of who asked for what, and the account and its history belong to whoever signed up, which becomes a problem the day they leave.

3. Whether customers can reach it on your channels

Manus is an agent the account holder instructs, through its apps and, in Cue's early access, through personal agents with their own phone numbers and email addresses that act for you. Your customers messaging your business number is a different shape. In Singapore that number is usually on WhatsApp, and what a business needs there is an agent that answers within an approved scope, hands over to staff when it should, and speaks as the company rather than as an employee. Test any agent the way a customer would before assuming it does that job.

4. Approvals and an audit trail for actions taken on the company's behalf

An agent that sends email, posts, books or pays is acting for the company, and Cue agents can pay within a budget the user sets. Before that touches company accounts, establish what waits for a human, who may approve it, and where each action is recorded. IMDA's Model AI Governance Framework for Agentic AI, voluntary guidance published in January 2026, is explicit that humans remain accountable for what these systems do.

Can you use Manus AI for free?

Yes, within limits. Manus's help centre describes a free plan with 300 credits refreshed daily, reset at midnight UTC, not carried over, capped each month and limited to the lighter mode. Paid use is sold as monthly credit bundles: the membership pricing article says the former Basic and Plus plans are now called Pro, and Team starts at US$20 per seat per month. Credits used depend on the task, so trial it on real work. Checked 16 October 2026; manus.im is authoritative.

Personal agent subscription or deployed business system?

As with Grok Bot, the dividing line is not capability. It is whose agent it is.

Scroll the table sideways to see every column.

When is Manus the right choice?

When the work belongs to a person. Manus fits if one person or a small team produces research, analysis, decks or prototypes; if the material is public or your own rather than customers' personal data; if you want results this week with nothing to deploy; or if a defined group of staff each need their own agent, which is what the Team plan is for. A deployed system would be overkill there. It stops fitting when customers are on the other end, when work runs on company channels, when several people depend on one workflow, or when actions need an approval and a record.

How Olano helps

Olano is a Singapore AI company, and Olano Studio, the team behind this site, designs, deploys and manages agent systems that belong to a business rather than to one person:

• On the company's channels. Agents answer customers on your WhatsApp number, Telegram, email and the rest of 15+ messaging channels; staff use the same agents from the dashboard.
• Isolated per client. An isolated deployment rather than a shared tenant, with custom AWS or GCP regions where residency matters.
• Approvals and an audit trail. Trust levels 0-4 and per-category approval gates; anything outbound waits for a person by default, and every action and approval is recorded.
• Any model per agent. Models from 18+ providers, mixed per agent, with your own API keys if you prefer.
• Priced per deployment, not per seat. Adding a colleague does not add a licence.

The limit, plainly: a deployment starts with a consultation and a fixed proposal and goes live once the workflow is built and tested, so it does not replace a personal agent you can start this afternoon. See AI agents in Singapore for how a deployed system works, our services for how an engagement runs, and why Olano is based in Singapore.