AI 模型和平台
聪明的实验室发布 399 光环 AI 眼镜带有记忆和适应性游戏
在 Google 上的首选来源中添加 Unite.AI
聪明的实验室正在给它的光环智能眼镜带来持续的AI记忆和适应性游戏，将399元可穿戴作为提问的方法，创造小的应用，并将日常体验转化为不断发展的游戏。 公司10月27日的公告将Noa的助手和一种叫做Narrative的游戏体验置于该方法的中心。
这个提议超越了把一个聊天员放在一副眼镜里。 聪明的实验室想要一个穿戴的，在互动之间搭载上下文，并用它来让未来的对话，工具，玩得更个人化。 它的光环产品页面将眼镜列为现在的货运。 公共硬件文件和固件也提供了公司如何建立该系统的更仔细的观察——其隐私政策对本地处理和云辅助AI作了重要的区分。
从AI助手到持久游戏
优秀的实验室将经验分为三项活动：问、做和玩。 问诺阿中心，一个对话助手，意在回忆喜好和早期的互动。 称 Noa公司没有要求用户用每个问题来重建相同的上下文，而是可以向前推进。
把这个想法推广到语音创建的小应用程序和工作流程中。 一个用户描述了他们想要建立的经验，而该平台旨在将这个请求变成一个小型的应用程序，可以通过Halo社区的应用商店共享。 吸引力是一条从一个想法到现在使用的工具的较短路径，而穿戴者不必写代码。
游戏是宣布方向更不寻常的地方。 叙述性被描述为一种适应性游戏，将现实世界的行动，发现和决定纳入持续的经验。 游戏与其遵循工作室设计的固定序列，不如打算与玩家一同改变。
这是公司的拟议经验，而不是这项公告所确定的业绩结果。 有意义的检验标准将是，持续的环境是否随着时间推移而产生可喜的、连贯的作用。 记住互动只是第一步；游戏也需要进步、可以理解的后果和返回的理由。
光环背后的硬件
发布规格包括全彩微OLED显示器，麦克风，骨导声器，蓝牙5.3，重量刚超过40克。 精明的实验室声称长达14小时的电池寿命；实际耐力将取决于使用。
该公司的硬件文件确定了一个Alif Balletto B1处理器，其中有一个Arm Cortex-M55 CPU和一个Ethos-U55神经处理器。 后者为支助的当地人工智能工作量提供专用硬件。 低分辨率的640乘以480Global-shutter色感应器可以使系统进行视觉输入，而双麦克风和立体骨导音频支持互动则没有手持屏幕。
这种组合说明了环境AI背后的工程权衡：一个磨损在面部的装置，电池，计算和热散的空间有限。 本地推论硬件可以高效地处理选定的任务，但其存在并不能确定整个助理运行离线。
隐私声称在实践中的平均值
聪明的实验室形容光环是隐私第一。 其发布释放的称视觉图像被处理端侧并丢弃，将传感器设定为上下文来源而不是常规摄影工具。
公司隐私政策于10月7日更新，称 Noa使用云AI进行交谈和图像理解，经许可后与Google的Gemini服务共享相关信息。 它还在加密云存储中保留称对话记录和相关内存数据，直到用户请求删除。
这些是产品数据流动的不同部分。 眼镜上的局部处理不应被理解为保证每个Noa特性都离线运行，或者任何视觉信息都达不到云服务。 对于一个基于内存的助手来说，相关问题包括哪些功能被启用，哪些权限被授予，以及如何删除保留的信息。
一个平台开发者可以检查和扩展
Halo的开发者吸引力部分取决于在应用层以下工作的能力。 辉煌实验室发布环绕Zephyr RTOS建造的卤素固件存储器，其中有一个用于端侧应用的Lua运行时段和蓝牙固件更新。 它的开发商资源也指向SDK和文档。
这使得开发商可以进行可穿戴界面、传感器和应用程序的实验。 寄存器包括具有不同许可证条款的组件，因此源代码的可用性应当与每个依赖都持有相同无限制许可证的说法区分开来。
对于更广泛的智能眼镜市场来说，这种方法使得光环作为一个可扩展的平台变得有趣。 语音生成的微型app可以让消费者方便地定制，而出版的固件则为技术用户提供了更深入地积累自身经验的路径。 两层面向不同的受众，但服务于同一个目标：让产品演进超越其最初的特征设定。
可穿戴AI的定价和下一个测试
光环造价399元。 Noa和迷你app包含有日常使用上限，产品页区分免费标准助理与付费对话级。 Noa Plus的每月定价为9.99美元，Noa Pro的每月定价为39.99美元，使目前的服务水平成为购买决定的一部分。
Goodwater Capital也支持Greatwater Labs
光环带来在消费者AI中结合了几个方向：持续记忆，自然语言的应用创建，可穿戴的计算，以及适应性的娱乐。 它的机会在于使这些能力发挥足够的作用，成为日常生活的一部分。 提供这种经验既需要可靠的互动，也需要围绕个人背景的明确界限，从而使这些经验成为可能。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
AI Models & Platforms
Brilliant Labs Launches $399 Halo AI Glasses With Memory and Adaptive Gaming
Add Unite.AI to your preferred sources on Google
Brilliant Labs is bringing persistent AI memory and adaptive gaming to its Halo smartglasses, positioning the $399 wearable as a way to ask questions, create small applications, and turn everyday experiences into an evolving game. The company’s October 27 announcement puts its Noa assistant and a game experience called Narrative at the center of that approach.
The proposition goes beyond putting a chatbot in a pair of glasses. Brilliant Labs wants a wearable that carries context between interactions and uses it to make future conversations, tools, and play more personal. Its Halo product page lists the glasses as shipping now. Public hardware documentation and firmware also offer a closer look at how the company is building the system—and its privacy policy makes an important distinction between local processing and cloud-assisted AI.
From an AI Assistant to a Persistent Game
Brilliant Labs groups the experience into three activities: Ask, Make, and Play. Ask centers on Noa, a conversational assistant intended to remember preferences and earlier interactions. Instead of requiring users to reconstruct the same context with every question, the company says Noa can carry it forward.
Make extends that idea into voice-created miniapps and workflows. A user describes an experience they want to build, and the platform is designed to turn that request into a small application that can be shared through the Halo community’s app store. The attraction is a shorter path from an idea to a tool used in the moment, without the wearer having to write code.
Play is where the announcement takes a more unusual direction. Narrative is described as an adaptive game that incorporates real-world actions, discoveries, and decisions into an ongoing experience. Rather than following a fixed sequence designed by a studio, the game is intended to change with the player.
That is the company’s proposed experience, rather than a performance result established by this announcement. The meaningful test will be whether persistent context produces enjoyable, coherent play over time. Remembering an interaction is only the first step; a game also needs progression, understandable consequences, and reasons to return.
The Hardware Behind Halo
The launch specifications include a full-color microOLED display, microphones, bone conduction speakers, Bluetooth 5.3, and a weight just over 40 grams. Brilliant Labs claims up to 14 hours of battery life; actual endurance will depend on use.
The company’s hardware documentation identifies an Alif Balletto B1 processor with an Arm Cortex-M55 CPU and an Ethos-U55 neural processing unit. The latter provides dedicated hardware for supported local AI workloads. A low-resolution 640-by-480 global-shutter color sensor gives the system visual input, while dual microphones and stereo bone conduction audio support interaction without a handheld screen.
This combination illustrates the engineering tradeoff behind ambient AI: a device worn on the face has limited room for batteries, computing, and heat dissipation. Local inference hardware can handle selected tasks efficiently, but its presence does not establish that the entire assistant runs offline.
What the Privacy Claims Mean in Practice
Brilliant Labs describes Halo as privacy-first. Its launch release says visual images are processed on-device and discarded, framing the sensor as a source of context rather than a conventional photography tool.
The company’s privacy policy, updated October 7, describes a broader cloud-assisted service. It says Noa uses cloud AI for conversation and image understanding and, with permission, shares relevant information with Google’s Gemini services. It also says conversation transcripts and associated memory data are retained in encrypted cloud storage until the user requests deletion.
These are different parts of the product’s data flow. Local processing on the glasses should not be read as a promise that every Noa feature operates offline or that no visual information can reach a cloud service. For a memory-based assistant, the relevant questions include which features are enabled, what permissions have been granted, and how retained information is deleted.
A Platform Developers Can Inspect and Extend
Halo’s developer appeal rests partly on the ability to work below the application layer. Brilliant Labs publishes a Halo firmware repository built around Zephyr RTOS, with a Lua runtime for on-device applications and Bluetooth firmware updates. Its developer resources also point to SDKs and documentation.
That gives developers concrete building blocks for experimenting with wearable interfaces, sensors, and applications. The repository includes components with different licensing terms, so the availability of source code should be distinguished from a claim that every dependency carries the same unrestricted license.
For the wider smartglasses market, this approach makes Halo interesting as an extensible platform. Voice-generated miniapps could make customization accessible to consumers, while published firmware gives technical users a deeper route to building their own experiences. The two layers address different audiences but serve the same goal: letting the product evolve beyond its original feature set.
Pricing and the Next Test for Wearable AI
Halo costs $399. Noa and miniapps are included with daily usage caps, and the product page distinguishes the free Standard assistant from paid conversational tiers. Monthly pricing is $9.99 for Noa Plus and $39.99 for Noa Pro, making the ongoing service level part of the buying decision.
Brilliant Labs also announced backing from Goodwater Capital, with participation from Brendan Iribe and other investors, without disclosing a funding amount in the release.
Halo brings together several directions in consumer AI: persistent memory, natural-language app creation, wearable computing, and adaptive entertainment. Its opportunity lies in making those capabilities useful enough to become part of daily life. Delivering that experience will require both dependable interactions and clear boundaries around the personal context that makes them possible.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.