Manus Project Initiation Core Discussion Minutes (PDF file)

The minutes of the initial meetings at the Manus project's inception meticulously document the team's in-depth discussions regarding product positioning, technical architecture, user experience, and human-computer collaboration models. Core issues included choosing a general-purpose approach.intelligentbodyOr is it a strategic choice for vertical domain experts to build a cloud environment with "agent" capabilities, design a user interface that balances trust and control, and explore... Agent As an extension of human intellect, the team established a strategy of "general-purpose platform + high-frequency scenario optimization," clarified key technical directions such as state persistence and cloud browsers, and proposed design concepts such as progressive disclosure, laying a solid foundation for the project's subsequent development.

Get theManus Project Initiation Core Discussion MinutesThe original PDF report is available by scanning the QR code and replying. 20251229

The Manus project's initial discussion minutes are compiled from recordings of two core meetings. Compared to the initial version, the expanded version more comprehensively recreates the discussion details, supplementing them with rich deliberation processes and analogies, providing a more valuable document for the project's subsequent development. The minutes cover in-depth discussions on various aspects, including product philosophy, technical architecture, user experience, and market strategy.

Discussion participants

Red, Peak, Fan Bin, Zhang Tao (hidecloud), Pan Pan (PanPan)

The discussion revolved around the following four aspects:

• Product Philosophy: Exploring how Manus should be in the "general"intelligentbodyPositioning itself between "vertical domain experts" and "Baidu model," it proposed an analogy between the "Baidu model" and the "Hao123 model," ultimately establishing a strategic direction of "prioritizing general applicability and gradually optimizing high-frequency scenarios."
• Technical ArchitectureThe research focuses on how to build a cloud environment with "proxy" capabilities, with particular attention to the implementation path of "cloud browser", the issue of "state persistence" across sessions, and the balance between security and ease of use.
• User experienceThe discussion focused on how to design product interfaces to meet the needs of different user groups, including managers who "only look at results" and engineers who "care about the process," and proposed the design concepts of "progressive disclosure" and "operating system metaphor."
• Human-machine collaboration mode: Exploration Agent The value of this lies in overcoming the limitations of human cognition and in clarifying the relationship between humans and... Agent An ideal collaboration model.

Product Philosophy: The Strategic Choice Between Generality and Vertical Optimization

• A comparison between the Baidu model and the Hao123 model:Red proposed,Agent The development path can be compared to the "Baidu model" and the "Hao123 model." The Hao123 model is similar to a traditional chatbot, where developers pre-integrate specific functions, limiting user capabilities and slowing expansion. The Baidu model, on the other hand, builds a general-purpose platform, optimizing through analysis of high-frequency tasks and launching "pre-defined capabilities." The team ultimately agreed on a "generality-first" strategy, believing that generality is the foundation for attracting users, while subsequent optimization is the key to building competitiveness.
• Boundaries and Challenges of UniversalityThe discussion pointed out that universality is a core strategy, but boundaries and challenges cannot be ignored. Fan Bin proposed that universality... Agent When facing professional software (such as video editing software), it will be difficult to compete with professional software in the short term. Peak suggests that if... Agent The ability to run in a virtual machine with a desktop environment, simulating keyboard and mouse operations to use professional software, might overcome the limitations. Furthermore, Red points out that general... Agent There may be issues with knowledge conflicts, and a "domain isolation" mechanism may be needed in the future.

Technical Architecture: Building a True "Cloud Proxy"

• Cloud browsers and remote interactionThe team explored the concept of "Browser in Browser" and studied XPRA. open sourceThe project provides a technical reference for low-latency interaction by streaming remote application interfaces. Zhang Tao (hidecloud) shared details of the project, believing it meets Manus's needs.
• State persistence problem:current Agent The product's "one-time session" mechanism is a core pain point. Manus needs to achieve comprehensive state persistence, including login state, file system, and key management. The team believes that persistence is the foundation for achieving true "proxy" functionality and can significantly improve the user experience.
• User takeover mechanism: When Agent When encountering complex CAPTCHAs or two-step verification, users need to be able to take over the browser to complete the operation. This mechanism is to compensate for... AI The key to ensuring the successful completion of the task despite insufficient capabilities.

User Interface and Interaction Experience: Finding a Balance Between "Trust" and "Control"

• The dual role of the interfaceDevin's interface design is divided into two columns: the left column is the conversation flow, and the right column is... Agent The workspace. This design satisfies the manager's need for trust and the engineer's need for control. Red stated that although he doesn't pay much attention to the right-hand window, its very existence builds trust.
• Criticism and Improvements to the Devin InterfaceThe team pointed out that the Devin interface suffered from information overload, a lack of overall overview, and disorganized functionality. Based on this, Manus proposed the design principles of "progressive disclosure" and "operating system metaphor" to create a simpler, more intuitive interface.High efficiencyThe user interface.

Human-computer collaboration:Agent As an extension of the human mind

• Overcoming the limitations of human cognition:Agent It overcomes the pitfalls of human empiricism and the lack of persistence in addressing problems, always starting from first principles to find the optimal solution. PanPan and Zhang Tao believe this is... Agent That is its core value.
• Complex Systems and Long-Term PlanningTaking the game EVE Online as an example,Agent It can act as the "general scheduler" of complex systems, helping humans manage and optimize massive projects.

After in-depth discussions, the team established the following core principles:

• Strategic levelIt adopts a dual-drive strategy of "general-purpose platform + high-frequency scenario optimization".
• Technical aspectsThe core of this architecture is "state persistence" and "cloud browser". It is designed to build a technical architecture with "proxy" capabilities.
• Product levelGuided by the principles of "progressive disclosure" and the "operating system metaphor," we aim to create a next-generation system that combines trust and a sense of control. Agent interface.

Following the discussion, the team acted swiftly, forming a project group, sharing preliminary data, and clarifying responsibilities. The Manus project officially launched, redefining...intelligentbodyTo become humanpowerfulThe extension of the mind.

Get theManus Project Initiation Core Discussion MinutesThe original PDF report is available by scanning the QR code and replying. 20251229