In a recent global strategy launch themed "AI in All," SIGENERGY unveiled SigenAgent, described as the energy sector's first comprehensive AI agent. Its operational logic involves users setting goals, the AI formulating plans, and devices executing tasks. It comprehensively interprets user objectives, analyzes real-time data on weather, electricity prices, grid conditions, and equipment status, devises optimal strategies through intelligent algorithms, and coordinates device execution, continuously tracking results and making dynamic adjustments—functioning like an autonomous, self-evolving "professional energy manager."
SIGENERGY's Chairman and CEO, Xu Yingtong, stated that the company aims to genuinely integrate AI with energy, making new attempts and progress each year in the "AI+Energy" field.
"I have our system installed in my own home. After connecting the personal assistant Hermes to the system, it unlocked unprecedented capabilities," Xu shared. He described feeling that the system had evolved to a new level, marking a distinct new phase in AI+Energy. "It is no longer just a voice or smart customer service from the past, but an assistant by your side that continuously evolves, truly understands you, and can autonomously manage the system. You simply tell it your goal via voice, and it executes automatically."
SigenAgent introduces four major vertical agents, transforming complex professional operations into clear goal-oriented instructions. AI decision-making relies heavily on predictions for electricity prices, photovoltaic power output, and load forecasting. The accuracy of these predictions depends on vast amounts of foundational data. How does SigenAgent address this?
Xu Yingtong explained that SIGENERGY's hardware devices, such as sensors and various interface panels, inherently possess robust data collection capabilities. For instance, using a DC bus solution, each mounted device includes numerous integrated sensors. Data from each interface is collected at millisecond-level intervals, with synchronization being a key challenge that has been resolved. After computation, only valuable data is uploaded to the cloud database for storage. When the AI performs scheduling, it only extracts data relevant to the specific task. This data has been accumulated over a long period.
He pointed out that while the underlying decision-making principles of past AI scheduling models were similar, previous orchestration relied more on generalized strategies, akin to tens of thousands of global devices sharing one "brain." The core upgrade now is that each user has a dedicated Agent with persistent memory, evolving alongside the user. This is the fundamental difference between a self-developed dedicated Agent and a general-purpose Q&A AI—shifting from a "shared brain" to personalized intelligent service characterized by "unique for each user, with persistent memory."
"In terms of data itself, we have deployed devices in over 80 countries. Regarding the number of online sites fully managed by AI, while there's no authoritative industry data for comparison, we believe we have the most online site data in the industry. These data sources are already sufficiently rich," Xu said.
Introducing intelligence into the core of energy control prioritizes safety and trust. SIGENERGY has established four core principles: User Sovereignty (critical operations require user confirmation), Data Compliance (with six regional cloud nodes globally for localized data storage), Offline Safety (capable of local operation without network connectivity), and Operational Transparency (all scheduling strategies and logic are clear and traceable).
Regarding managing risks associated with large language model (LLM) "hallucinations," Xu Yingtong stated that the value of LLMs lies in general data from the internet or other products, serving only as a reference in decision-making. LLMs primarily handle common-sense capabilities like language understanding, multilingual translation, and complex task orchestration. However, for specific device operations, control resides entirely with SIGENERGY's self-developed Agent, as the truly specialized data originates from SIGENERGY itself, an area where LLMs lack domain knowledge.
He discussed safety baselines, noting that all operational processes are fully transparent to the user. LLMs have a "black box" issue where their decision-making process is often unclear. However, for each critical operational step, SIGENERGY's self-developed Agent conducts safety checks and requires final user confirmation.
"Whether for individual or commercial/industrial users, the final decision-making authority currently remains with the user, not entirely handed over to AI," Xu stated. He mentioned that SIGENERGY will launch a beta version in Europe, particularly in regions with more dynamic pricing, with full availability expected in about a month. For developed markets like Europe and the US, SIGENERGY has established a complete technical system covering network security, GDPR compliance, and privacy protection.
At the launch event, SIGENERGY, in collaboration with Frost & Sullivan, also released the "2026 AI+New Energy White Paper" and innovatively proposed the Energy Intelligence Level (EIL) five-tier grading system. This system, referencing the logic of autonomous driving classification, aims to foster industry consensus on intelligence and accelerate the transition from single-point device intelligence to comprehensive system intelligence.
Xu Yingtong explained that in formulating this grading system, SIGENERGY referenced existing industry practices, such as autonomous driving intelligence levels and relatively fair systems in the AI field, combining general capabilities with the specific characteristics of the energy sector, like requirements for network infrastructure hardware, sensors, and control capabilities. Prior to this, there may have been no such classification standard proposed or attempted in the industry in this direction.
Xu noted that the current integration of AI and energy has a practical application foundation, enabling the implementation of this system. "I believe more and more companies will follow suit, allowing the industry to communicate under the same standard or discourse system, for example, clearly stating that a certain system has reached L2 or L3 level," he said. He admitted the current version is 1.0 and hopes industry peers will participate in its refinement.
Enterprises are facing challenges with rising token costs, as various AI services have begun commercial charging models. Will SigenAgent also charge fees?
In response, Xu Yingtong stated that SigenAgent is currently free for users. The infrastructure is hosted on cloud service providers like Amazon and Alibaba, and tokens need to be purchased from LLM providers such as DeepSeek and GPT, but future strategies to address this cost challenge are under consideration.
Regarding token costs and business models, Xu believes the business model of tokens itself is changing. Relying solely on expensive US LLMs like GPT or Claude makes widespread adoption unfeasible. With the development of Chinese LLM companies, especially the improving capabilities of many open-source models, the landscape is shifting. He suggested that LLM tokens might gradually evolve into a foundational service.
On the charging model, Xu revealed that, analogous to current network data plans for devices (like Wi-Fi or 4G ports), SigenAgent comes with a certain data package upon purchase. The future token charging model might be similar, becoming a basic service. He noted that all current SigenAgent-enabled devices shipped have full capabilities, fully open without feature tiering. Whether a tiered model is considered in the future will depend on technological developments.
"We still hope to rely primarily on hardware devices for profitability. But the related capabilities can be launched at any time—our marketplace already has data packages, and each app has corresponding functions," Xu said. He indicated that this is currently just a business decision issue, with no specific charging plans at present.