SentiCat, the AI Agent product under SentiPulse (Mind Spectrum), officially launched its public beta on April 23. The product is core-positioned as "task execution + emotional companionship", which adds long-term interaction and emotional companionship capabilities on the basis of task execution.
Public Beta Rules
- Application method: Queue for reservation on the official website, experience codes are issued after review.
- Quota: Users who pass official review get an initial 100 RMB Token quota, with subsequent top-ups based on usage.
- System requirements: Currently only supports Windows.
Technical Highlights
- Self-developed architecture: The system architecture is independently built by the team, not a secondary encapsulation based on existing Agent frameworks such as OpenClaw, covering the full chain from context management, task execution to memory system.
- Multi-model access: Simultaneously connects to 7 mainstream domestic large models including MiniMax, Tongyi Qianwen, Zhipu, Doubao, DeepSeek, Kimi, and Xiaomi, supports manual switching in settings, and is not tied to a single base model.
- Extensibility: Supports installing Skills and connecting to external MCP servers, providing system capabilities such as scheduled tasks, custom scripts (scripts generated from natural language), AI image generation, etc.
- Initialization process: There is an onboarding step for first-time installation, after authorization, a user profile is generated for preference adaptation in subsequent interactions.
Three Core Application Scenarios
- Office automation: Drives automatic PPT generation, Excel data processing and Word document formatting via natural language.
- In-depth research: After inputting a topic, the system automatically completes multi-source data integration, information collection and in-depth analysis, and delivers research reports.
- Code development: Covers the full development chain including code writing, bug locating and project refactoring.
Data Security
Adopts the local-first design principle: chat records, personal preferences and file materials are all stored on the user's device, and are not uploaded to the cloud by default. API keys are encrypted, and sandbox mechanisms are introduced for code execution and plugin operation. Only necessary information is sent when calling cloud models, and users can view, manage and delete local data at any time.
Link: Visit the SentiCat official website for details and to queue for application.