What changed
Manus says it is resuming operations as an independent company and is applying a data transition in specific jurisdictions. Its official service-change notice says that data generated by certain users between December 29, 2025 and 7:59 a.m. on August 23, 2026 (Singapore Time) is subject to deletion from 8:00 a.m. on August 23 through 7:59 a.m. on August 25, 2026 (SGT).
The notice distinguishes between affected and unaffected accounts. The in-app notice and account category, rather than a general location guess, determine the impact. Some accounts retain their account and subscription while losing post-acquisition task data and connector authorizations; Type C accounts may have account, subscription, and associated data deleted. Team accounts can be affected through the owner’s and members’ account categories.
Timeline readers should know
- Backup window: affected users can use Manus’s Data Backup Tool through 7:59 a.m. on August 23, 2026 (SGT). Data created or updated after a completed backup must be backed up again.
- Deletion window: 8:00 a.m. August 23 to 7:59 a.m. August 25, 2026 (SGT). Specified account information, subscription records, task data, generated artifacts, and connector authorizations may be permanently deleted.
- Restoration: Manus says the restoration portal opens at 8:00 a.m. on August 25, 2026 (SGT). Saved backup files are required to restore deleted task data and artifacts.
What affected users should do
- Check the in-app notice for each personal and team account; do not infer eligibility from country alone.
- Export account information and task data with the official backup tool before the deadline shown by Manus.
- For teams, have the owner back up both account and task data and coordinate with members.
- Keep backup files safe and verify that the export contains the artifacts and connector information you actually need.
- Review subscription cancellation and refund notices separately; these are account-specific and are not a substitute for a data backup.
Our editorial assessment
This is an operational and privacy notice, not a model-quality release. We are not treating the deletion policy as evidence that Manus is unsafe or that every user is affected. The practical conclusion is narrower: Manus users should check their official notice and maintain an independent copy of important work before relying on the service.
Primary sources: Manus official notice and Manus Help Center service-change overview.
Our editorial read
A useful agent-style option for multi-step research and execution, but it needs explicit boundaries.
Best fit
Users who can define a deliverable, provide sources and review an agent’s intermediate work before acting on it.
Where we would hesitate
Do not delegate irreversible actions, confidential uploads or high-stakes conclusions without human checkpoints.
Decision rule
Choose Manus when orchestration is the bottleneck. Choose a conventional research workflow when provenance and repeatability must be visible at every step.
Evidence note: This judgment synthesises the linked official documentation, release notes and product context. It is not a claim that AI Tools Manual independently benchmarked every tool.
September 28, 2026: Manus 2.0 is a broader workspace redesign
Manus describes 2.0 as a new architecture and product set rather than a routine version bump. Its Cascade agent harness selectively brings in specialist capabilities, while Manus Studio expands the desktop workspace for documents, spreadsheets, PDFs, slides, websites, code, games, and video. The most operationally significant change is Automations: alongside scheduled tasks, a connected-service event—such as a new email, calendar event, Slack message, or Notion update—can trigger a workflow. Manus also offers a purchasable Cloud Computer for projects that need a persistent environment. These features move Manus from one-off task execution toward workflows that can continue while a user is away; teams should therefore treat trigger design, connected-account scope, and human approval as core setup, not optional polish.
Cue is a separate early-access agent product
Manus says its new Cue app is available on web and desktop, with iOS pending App Store review. It is in early access and invite-based, so readers should not mistake the launch for general availability. The official description gives each Cue agent its own identity and digital resources, including email, phone number, wallet, and computer, and says users can set a spending budget. That model creates useful delegation but also material risks: an agent may send communications, make purchases within its configured budget, or act through connected services. Begin with a low-value sandbox task, use the narrowest permissions, set a small explicit budget, and require approval for external commitments. Do not connect production credentials until auditability and recovery procedures are understood.
Our assessment and source disclosure
Manus may suit founders and small teams that want one environment to move from research to deliverables and event-based follow-up. It is a weaker fit when workflows require strict infrastructure control, mature access governance, or a clear separation between drafting and acting. Manus reports that one Cascade configuration used 23.2% fewer tokens, finished 28.2% faster, and cost 32% less than its previous system; those are vendor results from a described configuration, not independently reproduced comparisons. Primary source: Manus’s Manus 2.0 announcement. Compare the automation model with Grok Bot and Perplexity Computer. AI Tools Manual did not test Manus 2.0 or verify privacy, payment, or safety controls; verify these details and current charges before approval.
September 29, 2026: Manus Flex brings supported model providers into the workspace
Manus announced Manus Flex as a module for using a user’s own API key from a supported inference provider inside the Manus workspace. The vendor says the key connects that provider’s inference to existing Manus projects, tools, execution environments, automations, and agent workflows. Manus continues to supply the agent harness and infrastructure, while users choose a supported provider, a primary model, and a reasoning-effort setting. The announcement names OpenRouter, Fireworks, and Modal as initial inference partners.
The billing boundary is split: Manus says model inference is billed directly by the connected provider, while other services and infrastructure used in a Manus task continue to consume Manus credits. This is a meaningful change for builders who already maintain a provider account and want more model choice without leaving the Manus agent workflow. It is not automatically a lower-cost option: total spend depends on provider token usage plus any Manus services consumed by the task.
Who should consider Flex—and what remains unverified
Flex may suit developers and technical teams that already manage provider accounts, need to select a particular model, and can monitor two billing surfaces. It is a weaker fit for occasional users who prefer one managed bill, or for teams that cannot approve a new place to configure credentials. The launch post does not document API-key storage or encryption, key-scoping controls, staff access, rotation and revocation behavior, retention in logs, supported regions, or how prompts and outputs are handled across Manus and the selected inference provider. We therefore cannot characterize the security or privacy of the key path from the announcement alone.
Before connecting a production key, verify the current provider/model list, key permissions and rotation process, the provider’s data-retention terms, and whether Manus exposes per-task usage or caps. Use a dedicated low-privilege key with a small spend limit, test on a non-sensitive project, and reconcile both the provider invoice and Manus credits. Review our AI tool privacy checklist and subscription-cost comparison guide before rollout.
Our assessment and evidence disclosure
Manus Flex expands choice and changes where model inference is billed, but the decision hinges on credential governance and end-to-end cost visibility—not merely on having more models. Keep secrets out of shared prompts and project files, use a dedicated provider key, and do not enable it for confidential work until the missing controls and data flows are confirmed. Primary source: Manus, “Introducing Manus Flex: Use your own API key in the Manus workspace,” September 29, 2026. AI Tools Manual did not connect a provider account or independently verify key security, data routing, usage accounting, availability, or plan eligibility.
October 1, 2026: Manus adds Game Dev and a layered Video Editor
Two official Manus posts extend the September 28 Manus 2.0 release. Game Dev adds a game-specific tweak panel for changing parameters such as speed, gravity, damage, and spawn rate while playing; it also supports managing and replacing game assets and building multiplayer experiences. Manus describes this as reducing the coding barrier, while acknowledging that Manus 2.0 still has rough edges. Video Editor is a separate creative workflow inside Manus Studio: it places generated footage, images, captions, motion graphics, voiceover, music, and sound effects on separate timeline layers that can be edited by hand or through instructions. The editor runs in the desktop app for macOS and Windows; the web app directs users to install the desktop app for the full timeline.
These additions broaden Manus beyond general research and office deliverables into interactive game prototyping and video assembly. Game Dev may fit a playable prototype or a small group experiment; it should not be treated as a replacement for professional game-engine workflows without checking export, source-code ownership, deployment, performance, and multiplayer controls. Video Editor may help creators turn prompts or local footage into an editable first cut, but the official post is a product-team account and not independent evidence of time savings or production quality. Creators should confirm supported file formats, asset rights, local-file handling, export resolution, and account eligibility before using it for client work.
Our assessment: These features make Manus more compelling when one workspace needs to move from an idea to a rough interactive or audiovisual artifact. The trade-off is that creative judgment and final verification still belong to the user, and the desktop dependency matters for anyone expecting browser-only editing. Try a non-sensitive sample project, preserve an exportable copy, and review each game mechanic or video fact before sharing. Primary sources: Manus, “Introducing Game Dev,” October 1, 2026 and Manus, “Introducing Video Editor in Manus 2.0,” October 1, 2026. The existing Manus Flex revision remains in human review for API-key custody, data routing, and billing; this addition does not resolve those open questions.
October 13, 2026: Manus agent capabilities are coming to OPPO phones
Manus announced a partnership with OPPO to integrate its agent capabilities into AI Mind Pilot’s Agent Mode on the upcoming OPPO Find X10 series with ColorOS 17. The partner announcement describes multi-step phone tasks involving files, photos and information across apps—for example, organizing travel details or preparing a presentation from existing materials. It says more details on supported markets, availability and user benefits will follow at launch. See Manus’s official announcement.
This is a device-integrated distribution path, not a general Manus mobile-app launch. It may help users who want an agent to work with phone context without manually uploading everything, but the initial device and OS are constrained. The announcement says ColorOS 17 uses on-device models for local file/photo organization and requests explicit confirmation before relevant content is uploaded for external processing. These are vendor-described design statements, not an independent privacy audit.
Before relying on it, confirm supported Find X10 variants and markets, which apps/data can be read, what actions require approval, what leaves the phone, retention and deletion behavior, and how the flow interacts with Manus and OPPO account terms. We have not tested the device experience or inspected network traffic. For desktop agent alternatives, compare Kimi Work and Lovable. Next review: November 16, 2026.
Human review required: verify launch timing, device/region availability, on-device versus cloud processing boundaries, consent prompts, retention, access scope and the official privacy terms before publication.