Terminus and BUTTONS Jointly Usher in a New Wave of Consumer-Grade Smart Entities

As a representative of AI service providers, Terminus has launched a new intelligent agent system—HALI. This system innovatively integrates end-to-end reinforcement learning technology with high-quality industry data. Not only does it break the limitations of traditional intelligent agents that rely on manual rule settings, but it also achieves automation in tool invocation, task decomposition, and multi-agent collaboration for the first time, significantly enhancing the agent's autonomous judgment capabilities, operational flexibility, and generalization abilities in complex scenarios.

In the deployment of consumer-grade intelligent agents, Terminus has partnered with the global intelligent luxury brand BUTTONS to apply the HALI system to its latest headphone series, BUTTONS CLIP. Some AI functions have been made available for user experience, creating a brand-new human-machine interaction model. In the future, the two parties will continue to collaborate in launching various AI products, including AR glass, smart ring, acoustic device, and smart home robot, jointly expanding the boundaries of smart hardware applications.

Human-like Thinking: Say Goodbye to the Shackles of "Manual Orchestration"

At present, most intelligent agents still rely on traditional workflow logic, splitting complex tasks and then integrating them. While this method is somewhat effective, it heavily depends on initial system settings and lacks flexibility, making it challenging to meet more diverse and complex task requirements.

To break through this limitation, Terminus innovatively adopts an end-to-end reinforcement learning approach, training the HALI model with high-quality data and driving it to continuously optimize strategies through interaction with the external environment using reward functions. The intelligent agent can autonomously reason, plan, and execute based on user instructions, completely getting rid of the dependence on manually defined workflows. Tool invocation, task decomposition, and self-verification can all be coherently completed within the model, enhancing the simplicity and accuracy of the overall system.

This underlying architecture innovation is not only a critical step in pushing AI models from an engineering level towards human-like intelligence but also lays the foundation for creating truly autonomously thinking intelligent agents.

Persistent Memory: A "Super Brain" that Balances Efficiency and Cost

Long-term memory capability is an important indicator of the maturity of intelligent agents. To achieve both real-time response and long-lasting memory, Terminus has introduced an efficient data compression and intelligent retrieval mechanism based on a knowledge graph in HALI.

Compared to traditional methods that rely on complete dialogue records and keyword retrieval, HALI can automatically extract the core information from user interactions, filter redundant content, greatly improve storage density, and compress system load. On this basis, by constructing user profiles and knowledge graphs, the system becomes more precise and efficient in understanding user intentions and outputting retrieval results, ensuring reasoning response speed while reducing storage and computational costs.

This solution not only alleviates the resource pressure brought by long-term memory but also makes personalized, companion-style AI possible.

Multi-Agent Collaboration: Breaking Through Latency Bottlenecks

Achieving coordination and task division among different intelligent agents is key to enhancing user experience. To this end, Terminus adopts a hybrid multi-model architecture, deploying fine-tuned small models on mobile terminals for low-latency response, while complex tasks are handled by server-side large models to ensure processing depth and accuracy.

This parallel solution effectively solves the problem of sluggish response in traditional systems, ensuring that users receive fast and accurate feedback whether issuing simple commands or complex requests.

Creating a New Intelligent Luxury Experience with Global Luxury Brand BUTTONS

Currently, the HALI system has achieved full connectivity between mobile and PC terminals. The world's first intelligent agent headphones, BUTTONS CLIP, equipped with this system have been officially launched, and some AI functions are open for users to experience.

This product uses aerospace-grade titanium metal and skin-friendly silicone materials, combining trendy aesthetics with the concept of affordable luxury, redefining the headphone experience. No longer confined to traditional competition dimensions of sound quality and noise reduction, BUTTONS CLIP is given a new position as a "mobile music living room". After the initial binding, users can enjoy seamless conversation and personalized AI services by simply awakening with the voice command "HALI HALI".

HALI can not only accurately understand instructions but also achieve habit learning and preference memory based on user behavior, providing truly personalized companionship and ushering in a new era of intelligent luxury.