In the past two years, many brands have launched AI PCs, but most of them are simply adding AI dialogue plugins to traditional computers. Essentially, it’s still a matter of humans directing software, requiring manual operation step by step. Users still need to manually open documents, switch software, and copy and paste to issue commands. AI is merely an auxiliary question-and-answer tool and cannot streamline the entire workflow.
At this year’s ChinaJoy 2026, Sixunited officially showcased its full range of Agent PC smart hardware. Unlike conventional AI laptops and mini PCs on the market, the Agent PC is positioned as an AI-native intelligent terminal. Its core concept completely reverses the human-machine relationship: users only need to state the final goal, and the computer autonomously breaks down the task, calls the program, continuously executes, and delivers the complete result.
As one of the few domestic manufacturers to achieve self-developed and integrated Agent capabilities across software and hardware, Sixlink Intelligent leverages its years of ODM hardware manufacturing capabilities, coupled with its self-developed SixClaw intelligent agent framework, AI Assistant 3.0, and EAM enterprise intelligent agent management platform, to launch an Agent PC product line covering mini-PCs, laptops, and all-in-ones. This time, we focus on the mini workstation version showcased at our booth, combining it with real-world office scenario testing to clarify whether the Agent PC is merely a conceptual gimmick or the next generation of productivity hardware.
I. Hardware Architecture: Tailor-made for Continuously Running AI Intelligent Agents
Ordinary PC hardware designs prioritize instantaneous performance for human use; however, Agent PCs require long-term background operation of large local models, vector knowledge bases, and concurrent tasks by multiple intelligent agents, necessitating targeted hardware optimization.
Computing Power Platform
The SixLink Intelligent Agent PC offers two mainstream chip options:
AMD Ryzen AI Max series platform: unified memory architecture, ample NPU computing power, suitable for lightweight local continuous inference of large models;
Intel Panther Lake platform, up to 180 TOPS of comprehensive AI computing power, targeting heavy enterprise knowledge base and batch document processing scenarios.
All models come standard with ultra-large memory expansion capabilities, supporting up to 128GB of memory. This is crucial for running local multi-agent systems and loading enterprise-owned document vector libraries. Ordinary thin and light AI PCs typically have a maximum of 64GB of memory, making it difficult to stably execute multiple automated workflows in parallel.
Form Factor and Expansion (Mini PC Version)
The chassis is only about 1.2L in size, with a unibody metal design and VESA wall mounting support, allowing direct mounting to the back of a monitor, significantly saving office space.
Dual M.2 PCIe 4.0 drive bays support up to 8TB of local solid-state storage for storing enterprise local knowledge bases and offline model files.
Rich video output interfaces support simultaneous output to four screens, meeting the needs of multi-window collaborative work for finance, design, and administration.
Low power consumption and continuous stable scheduling support long-term 24/7 standby operation of intelligent agent tasks.
From a hardware perspective, it’s clear that SixLink didn’t simply modify an existing Mini PC; instead, it restructured the entire machine’s power supply, heat dissipation, and memory scheduling strategies around “continuous local agent operation,” solving the pain point of ordinary computers easily throttling and lagging when running edge AI for extended periods.
II. Core Software Experience: SixClaw Intelligent Agent, Truly Achieving Autonomous Task Loop
Hardware is just the foundation; the self-developed software ecosystem is the biggest differentiator of the SixLink Agent PC. The entire system consists of three main modules: AI Assistant 3.0, SixClaw Intelligent Agent Engine, and Memo Long-Term Memory System.
Traditional PC vs. Agent PC Logic Comparison
Test Scenario 1: Routine Administrative Office Automation
Traditional AI PC Operation Process: Open meeting recording → Copy audio file → Start AI transcription → Manually organize key points → Create a new Word document to write minutes → Copy content to generate PPT.
At least 6 manual software switches are required throughout the process.
Six-link Agent PC Command (one sentence):
“Organize this morning’s department meeting recording into a structured meeting minutes, extract to-do items, automatically generate a report PPT, save it to the project folder, and send it via email to all members.”
Execution Process: The agent autonomously identifies local audio files, calls the speech model for transcription, distinguishes speakers, organizes the task list, launches office software to generate documents and slides, and then sends them in batches via email client. No manual intervention is required throughout the process.
Test Scenario 2: Enterprise Private Document Knowledge Base Retrieval
Many SMEs are concerned about the risk of data leakage when uploading internal contracts and plans to the cloud using AI.
Agent PC supports local private deployment of vector databases, storing all document data locally without uploading to third-party cloud services. We stored dozens of project contracts and solution documents locally and directly asked:
“Filter all partners with payment periods exceeding 60 days in their contracts and compile a comparison table.” The device autonomously traverses local files, extracts information, categorizes data, and automatically outputs an Excel report. Relying on a hybrid edge-cloud inference mechanism: confidential information is processed entirely locally; only open information query tasks connect to the network on demand, balancing security and practicality.
Three Core Software Highlights
Long-Term Memory Capability
The Memo memory module continuously records user work habits, document paths, and writing style. With long-term use, the intelligent agent becomes familiar with your work logic, reducing repetitive command descriptions.
Intelligent Agent Store, One-Click Workflow Deployment
Built-in scenario-based intelligent agent templates: contract review intelligent agent, meeting intelligent agent, material organization intelligent agent, code assistance intelligent agent. Ordinary users do not need to learn prompts and can directly enable pre-built workflows. Enterprises can also customize their own exclusive agents based on open interfaces.
EAM Enterprise-Level Management Platform
Designed for team deployment scenarios, this platform allows administrators to centrally manage permissions across multiple Agent PCs, synchronize knowledge bases, and approve tasks. It is suitable for bulk purchases by law firms, design institutes, and government/enterprise units.
III. Existing Shortcomings and Objective Limitations (Rational Evaluation)
Based on our on-site experience, we objectively summarize the product’s current shortcomings:
Currently targeting the B2B market, there is no retail version available for general individual consumers.
The SixLinks Intelligent Agent PC is primarily geared towards enterprise customization and bulk project purchases. There is currently no publicly available retail price, making it difficult for general individual users to purchase it individually in the short term.
Compatibility Boundaries Exist in Complex Cross-Software Workflows
The agent’s cross-program scheduling capabilities are limited for some niche industry-specific software. Currently, it is well-adapted to mainstream office software such as Office, WPS, Lark, and DingTalk.
Hardware Barriers Exist for Local Deployment of Large Models
For smooth operation of large models with parameters exceeding 70B, a configuration of 64GB or more of memory is recommended. Entry-level configurations are only suitable for lightweight office agents.
IV. Who Should Buy It? Target User Profile
✅ Recommended for Purchase
Small and medium-sized enterprises’ administrative, legal, and financial teams that frequently handle contracts, reports, and meeting materials, value data privacy, and are unwilling to upload internal documents to public clouds;
Studios, design, and consulting industries with large amounts of document organization and proposal writing work;
Government and enterprise units with private AI deployment needs seeking secure and controllable local AI hardware terminals;
AI developers need a long-term online local computing power terminal for continuous debugging of Agent workflows.
❌ Not Suitable
For ordinary individual users with only audio-visual entertainment and gaming needs, and who only require simple AI chat functions, ordinary consumer-grade AI PCs offer better value for money.
V. Evaluation Conclusion: Agent PC Opens the Next Generation of PC Competition
In the past two years, the AI PC competition has been limited to NPU computing power and benchmark parameters. However, the Six-Link Intelligent Agent PC proves that the next stage of PC evolution is no longer about the computing power of a single chip, but about the integrated hardware and software intelligent agent scheduling capabilities.
Simply adding an NPU and pre-installing AI software only qualifies as “a traditional computer with AI added”; the Agent PC redefines the terminal’s positioning—evolving from a “human-operated tool” to a “digital employee capable of autonomous work.”
Of course, the entire Agent PC market is still in its early stages, with significant room for growth in its ecosystem and software compatibility. However, Sixlink Intelligent’s complete implementation solution provides the industry with a clear direction: the core value of future office hardware lies in freeing humans from a large amount of repetitive, fragmented computer operations.
As more manufacturers follow suit with the Agent architecture, hardware and software integrated on-device AI hardware, represented by Sixlink Intelligent, may reshape the product standards for office computers over the next three years.