Manus Releases Manus 2.0, the Alchemy Video Editor, Game Dev, and Personal Agent Cue

Manus has released Manus 2.0. Alongside an upgrade to the underlying framework that orchestrates its Agents, the release introduces cloud computers, event automation, Manus Studio, a video editor, Alchemy, Game Dev, Computer Use, and Cue, a standalone personal Agent app.

Together, these features address a practical problem: earlier Agents were good at generating something once, but not at keeping the work going. After a video was generated, replacing only one segment was difficult. Automation stopped when the local computer shut down. Tasks could not start on their own when an email arrived, a calendar event began, or business data changed. Manus 2.0 aims to turn one-off tasks into projects that keep running, remain editable, and can be handed back to a person at any time.

What are the major features in Manus 2.0?

• Cascade: a new Agent framework that loads tools on demand and executes independent steps in parallel.
• Cloud computers: persistent computers for projects that can run automation, apps, game servers, and rendering jobs.
• Event automation: tasks can start not only on a schedule, but also when an email arrives, a cart is abandoned, a competitor changes a price, or a calendar event occurs.
• Manus Studio: documents, spreadsheets, slides, websites, code, games, and videos live in one project space.
• Video editor and Alchemy: generate a first cut, then edit the timeline, media, text, and audio directly. Alchemy combines video generation and code generation to create more complex work.
• Game Dev: start from playable game templates, manage assets and code, then publish to the web or connect a cloud computer to run a multiplayer server.
• Computer Use: remotely operate files, browsers, and applications on a local computer within the permissions the user grants.
• Cue: give different personal Agents their own email addresses, phone numbers, computers, spending limits, automations, and memories.

The official overview video below ties these capabilities together, moving from video generation and multiplayer games to cloud computers, event automation, remote control, and Cue's email, phone, and collaboration features.

Cascade: load each tool only when it is needed

Cascade is Manus's new in-house Agent framework. It is not a new large language model. It is the layer that prepares tools for the model, preserves project context, schedules execution steps, and handles the results.

Older Agents often loaded extensive tool instructions into the context before a task began, whether or not those tools were needed. Cascade keeps a project lightweight at first: it loads video capabilities only when the task calls for editing video, and a development environment only when code must be written. If searching for material and generating a voiceover do not depend on each other, it can also run them at the same time.

All three numbers point to the same idea: eliminate unnecessary context and waiting so the Agent can finish a task with fewer resources. They come from one test configuration run by Manus and are useful for understanding the direction of Cascade's optimization; they do not mean that every task will improve by a fixed 30%.

Cloud computers and event automation: the project keeps running after people go offline

For an Agent to keep working, it needs a computer that does not disappear when someone closes a laptop. Manus gives each project an independent cloud-computer environment where files, services, and processes can persist.

Event automation determines when a task should begin. It is no longer limited to instructions such as “run every day at 9:00.” It can listen to connected services: a new email, an abandoned Shopify cart, a change in advertising data, the start of a calendar event, or a competitor's price change can all trigger the next task.

The relationship is straightforward when the two are viewed together: event automation wakes the task, a cloud computer gives it somewhere to keep running, and Cascade brings in the tools it needs.

Manus Studio: AI delivers an editable project, not just a file

Manus Studio is the new unified creation space. Websites, apps, and automation live under Build; documents, slides, and media live under Create. A project can also begin with local files.

The most important change is not the number of tools, but the form of delivery. After AI completes the first pass, the project still retains its media, code, timeline, and context. A person can enter the project, revise the details, and then ask Manus to continue from the edited version.

Video editor: let AI make the first cut, then refine the details on the timeline

Earlier one-click video generation worked like a vending machine: a prompt went in and a finished video came out. Even replacing a single product image, changing the music, or shortening one shot could require regenerating the entire video.

The Manus video editor separates clips, images, text, stickers, motion graphics, and audio into individual assets, then places them on a timeline. An Agent can edit them, or a person can drag and replace them directly.

In the official demonstration, a user first creates a 15-second video from three coastal clips, then drags the final clip to shorten the total length to 13 seconds. The earlier material does not need to be described or generated again.

Alchemy: combine code, media, generation models, and editing around the goal

Alchemy is the video editor's high-quality creation mode. The user describes the intended work, and the system selects capabilities such as video generation, code rendering, media search, voiceover, and editing before returning the result to the timeline.

Manus Alchemy presents 12 examples of creating videos, each with a video and the full prompt. These prompts can also be given to Opus 5.5 or Sonnet 5.5 in an Agent environment that can execute code, render and export results, and invoke tools.

The library below preserves all 12 demo videos and provides the Chinese translation and original English text for each prompt. Select an example to view the corresponding work and copy the prompt.

The most useful lesson from these examples is not any fixed phrase, but how a prompt defines the objective, language, aspect ratio, duration, source of media, and delivery format at the same time. Equilibrium, for example, explicitly rules out generated video and stock footage, requires code rendering instead, and asks for an MP4, source code, and a compressed version for X. BEAT RUSH specifies generating a dancer with Seedance first, then creating rhythm-game effects in code.

A short prompt can establish the creative direction, but the demo still depends on the production environment behind it. In an Agent that lacks media search, code execution, TTS, video models, and rendering capabilities, the same prompt may produce only a script or a production plan.

Game Dev: from a playable template to publishing and multiplayer

Game Dev applies Studio's project model to game development. A user can begin with playable templates such as multiplayer or card games while managing characters, backgrounds, props, code, and the running game together.

If a game needs multiplayer support, it can connect to a cloud computer so the server stays online. Manus therefore covers more than “generating a piece of game code”: it also spans asset revisions, runtime debugging, web publishing, and server hosting.

Computer Use: operate an authorized computer from a phone

Computer Use lets Manus operate files, browsers, and applications on a local computer. The official example is someone away from home using a phone to ask Manus to find the newest presentation in a folder on their home computer and send it over, while watching the operation in real time.

This capability connects cloud projects with work that already exists locally. It does not create another copy of the files. It uses the existing computer environment in a visible, supervised session.

Cue: give each Agent its own identity and working resources

Cue is the personal Agent app that Manus launched at the same time. Each Agent can have its own email address, phone number, computer, spending limit, automations, Instructions, and Memory.

For example, one Agent might handle travel email and phone calls while another manages routine purchases. They can send messages, answer calls, leave summaries, and make payments within a budget set by the user. Cue aims to turn an Agent from a chat persona into a digital assistant with communication channels and a working environment.

From a single prompt to a project that keeps running

The eight capabilities in Manus 2.0 are not a collection of disconnected features. They follow one continuous workflow:

1. The user states a goal, and Cascade selects tools and schedules the steps;
2. Email, calendar, or business events can also trigger the task automatically;
3. A cloud computer keeps files, processes, and services online;
4. Studio preserves the project structure and all editable materials;
5. A person can edit the timeline, media, code, or documents;
6. Manus can continue from that version, publish it, or keep it running over time.

That is the core pain point this release addresses: AI should not only produce a surprising first version. It should also let the project keep running and remain editable after that first version, then hand control back to a person at the right moment.

What will determine the real experience is whether these long workflows run reliably, permissions are easy to control, tasks can recover after a failure, and the costs of cloud computers, video generation, and continuous automation are predictable. A feature count cannot answer those questions; they will need to be tested in real projects.