SuperAPI AI Directory
Partner with us中文AI newsFavorites
DirectoryNews and tutorials

StepAmoo - A New Generation Personal Agent Launched by StepFun

Author: AI小集 · Original source: AI工具集

What is StepAmoo

StepAmoo is a new generation personal agent launched by StepFun, built on the Step AOS agent-native operating system and possessing an operating-system-level identity. StepAmoo has five core capabilities: perception, memory, planning, connection, and execution. It adopts a dual-domain, three-step memory structure and can continuously understand user habits and preferences. Through end-cloud collaborative routing, StepAmoo can schedule atomic services across applications, complete complex task loops, and seamlessly continue across multiple devices.

StepAmoo

StepAmoo's Main Features

  • Multimodal Perception: Understands user intent and environmental status through natural means such as voice, vision, and text.
  • Dual-Domain Memory Accumulation: The user domain records personal information and preferences, while the agent domain accumulates vertical-domain knowledge and experience.
  • Autonomous Task Planning: Breaks down user intent into executable steps and autonomously formulates cross-application action plans.
  • Atomic Service Connection: Invokes system and application capabilities through the atomic capability engine to complete cross-service loops.
  • End-Cloud Collaborative Execution: Dynamically selects the device-side fast-response path or the cloud-side deep-reasoning path based on task complexity.
  • Result Interaction and Delivery: Users only need to express intent, and the agent completes the full chain and proactively feeds back results.
  • Four-Dimensional Security Governance: Trustworthy operations, visible processes, controllable permissions, and reversible behavior, fully auditable throughout.

StepAmoo's Technical Principles

  • Agent-Native System: Rebuilt based on Step AOS, allocating hardware resources to agents and scheduling system capabilities, rather than adding an AI entry point on top of a traditional OS.
  • Dual-Domain, Three-Step Memory: The user domain stores facts, episodes, and profiles, while the agent domain accumulates knowledge, cognition, and methods; through the "record-organize-recall" three-step chain, memory extraction, evolution, and retrieval are achieved.
  • End-Cloud Collaborative Routing: Dynamically switches among Fast Path (device-side loop), Deep Path (cloud-side reasoning), and Hybrid Path (end-cloud relay) based on task complexity, intent confidence, network status, and privacy level.
  • Cascaded Model Scheduling: Adopts the Step model Cascade architecture, with device-side Step Edge (300M-10B) responsible for real-time response, and cloud-side Step Flash (200B-400B) and Step Pro (1.5T+) handling complex reasoning.
  • Atomic Capability Engine: Breaks down system and application capabilities into minimal atomic units, opened through protocols such as MCP, A2A, and CLI for agents to freely orchestrate and invoke.
  • Unified Semantic Data Layer: Processes perception, behavior, and personal data into unified semantic files, and with a high-speed hybrid database enables extremely fast transmission and retrieval.

How to Use StepAmoo

StepAmoo is not yet publicly available and is in the early co-creation and internal testing stage.

StepAmoo's Core Advantages

  • Operating-System-Level Identity: Natively embedded in Step AOS, with system-level permissions, not an AI entry point added afterward.
  • Dual-Domain, Three-Step Memory: The user domain and agent domain are separated, accumulated through the record-organize-recall chain, with memory recall in as fast as 15 milliseconds.
  • End-Cloud Collaborative Routing: Dynamically switches between device-side fast response, cloud-side deep reasoning, or end-cloud relay mode based on task complexity.
  • Atomic Capability Orchestration: Decompose system and application services through the atomic capability engine, and freely orchestrate cross-application task loops.
  • Four-Dimensional Security Governance: Trustworthy, visible, controllable, and reversible, with the entire operation process auditable and permissions granted on demand and revoked once used.

Comparison of StepAmoo with Similar Competitors

Dimension StepAmoo Traditional Voice Assistants (such as Siri / Xiao Ai Tongxue)
System Identity Operating system-level native agent with system-level permissions Pre-installed application or feature entry point on the system, with limited permissions
Memory Capability Dual-domain three-step memory, accumulating user preferences and experience over the long term Short-term instruction memory, with no sustained contextual understanding
Cross-Application Orchestration Atomic service orchestration, capable of completing long-chain closed-loop tasks Limited invocation of specific Apps, making cross-application collaboration difficult
Device-Cloud Collaboration Dynamic routing relay between fast on-device responses and deep cloud inference Mainly relies on a single processing path, either cloud or device
Security Mechanism Four-dimensional security system, with auditable and revocable operations Basic permission management, lacking an agent-level governance framework

StepAmoo Application Scenarios

  • One-Stop Smart Travel: Users only need to state their destination and preferences, and StepAmoo automatically completes a cross-application loop of flight search, hotel booking, ride-hailing, and route navigation.
  • Cross-Application Office Efficiency: Links WPS, Calendar, and Email applications, automatically organizing to-do items from meeting minutes and syncing them to the schedule, achieving a seamless connection from document editing to task distribution.
  • Local Life Services: Based on user taste preferences and historical orders, automatically completes food delivery ordering on Meituan and plans a pickup route on Amap, handling life's trivial matters with one sentence.
  • Content Creation Assistance: Invokes creation tools such as CapCut, automatically searches album materials, generates video drafts, and publishes them to Weibo, connecting the complete chain from inspiration to publication.
  • Personalized Schedule Management: Identifies user habits based on long-term memory, proactively reminds about meetings at appropriate times, recommends today's outfit, and books commuting vehicles, becoming an all-weather personal assistant.