| 本文作者： Nemo | 2026-09-14 10:55 |
9月14日，豆包手机助手消费者版本正式发布。
豆包手机助手以豆包 App 为基础，与手机厂商在平台层面展开合作，围绕交互、任务执行、本地记忆及检索等核心感受，进一步探索 AI 在手机中的应用。据官方视频介绍，相比去年底推出的工艺预览版，这次的消费者版本更强调稳定性和日常可用性。
交互感受是此次豆包手机助手消费者版本重点介绍的基础能力之一。据介绍，豆包手机助手兼容语音、AI 键等多种唤醒方式，以适配不同使用情形。其中，专属 AI 键带有指纹鉴权能力，完成本人验证后无需二次解锁，即可调用助手执行相关任务。语音唤醒维度，豆包手机助手也针对远场、内噪和嘈杂环境等情形开展了优化。
另外，值得一提的是手机助手的显示屏问答能力。使用者无需额外截图或切换应用，即可让助手结合当前显示屏内容继续处理任务。例如，在官方视频演示中，使用者打开相机对准室内环境，说出“依据这个装修风格，帮我选一个一米二以内、售价不超过一千元的柜子”，豆包便可结合画面信息和使用者要求，在电商平台找到符合条件的商品建议。
类似的显示屏上下文理解能力，此前也曾浮现在苹果关于 Siri 和 Apple Intelligence 的演示中。这意味着，手机 AI 助手正从单纯回答难题，逐步转向理解使用者所处的具体情形和实际需要，并进一步完成相应任务。
除交互能力外，个人 Context 与记忆能力也是手机助手能否满足好使用者个性化需要的关键。据介绍，豆包手机助手为本地数据给出了搜索工具，可以在使用者需要时检索相册、短信、便签等信息。例如，当使用者询问港澳通行证的到期时间时，助手可以从手机中的相关信息里找到答案。
围绕办公情形，消费者版本还升级了录音感受，并接入飞书妙记能力。结合本地搜索工具，使用者可以直接追问具体内容，例如“昨天使用者访谈里提到的机型建议”。另外，使用者还可以借助语音输入、三指上滑，或同时按住 AI 键和音量键等方式，手动保存当前显示屏内容，形成可在本地检索的记忆。
任务执行是豆包手机助手此次展示的另一项核心能力。以出行情形为例，使用者说“帮我叫一辆去公司的车”，豆包即可确认常用地址、挑选车型并发起下单。
值得一提的是，工艺预览版发布后受到广泛关注的“操作手机”能力，此次以 Beta 版形式开放。豆包维度表示，图形界面智能体仍是一项快速发展的前沿能力，豆包手机助手期望与合作伙伴共同构建开放、规范的手机AI应用生态。为此，豆包手机助手推出显示屏自动化操作声明协议（SAEP，Screen Automation Execution Protocol），并同步启动为期30天的规则公示。依据该协议，第三方应用可以自主声明操作边界，明确允许或拒绝AI助手在本应用内开展显示屏自动化操作。豆包手机助手维度表示，将充分尊重开发者意愿，对明确拒绝相关操作的第三方应用不会开展自动化操作。
同时，豆包手机助手也在发布视频中重点介绍了隐私与安全的相关工作。据介绍，豆包手机助手在数据收集和处理上遵循“合理必要”与“使用者自主控制”原则，同时建立了 Agent 防护体系，对可操作范围开展分层、分级管理。发布视频还提到，使用者可以前往豆包手机助手官网查看完整的隐私安全白皮书。
据悉，首个豆包手机助手消费者版本将搭载在努比亚 NaviX Ultra 手机上，当下已启动预约，将于9月16日正式发售。
本文来源：https://www.leiphone.com/category/industrynews/p1zBbVMbsyYNHOJ4.html
若有版权问题，请来信告知：13951431913#139.com 或在右方提交评论。
Author: Nemo 2026-09-14 10:55
On September 14, the consumer version of Doubao Phone Assistant was officially released.
Based on the Doubao App, Doubao Phone Assistant collaborates with phone manufacturers at the platform level to further explore AI applications in phones, focusing on interaction, task execution, local memory, and retrieval. According to the official video, this consumer version emphasizes stability and daily usability compared to the craft preview version released at the end of last year.
Interaction is one of the core features highlighted in this consumer version. The assistant supports multiple wake-up methods, including voice and AI key, to adapt to different usage scenarios. The dedicated AI key features fingerprint authentication, allowing direct access to the assistant without a secondary unlock after identity verification. Voice wake-up has also been optimized for distant fields, internal noise, and noisy environments.
Another notable feature is the screen-based question-and-answer capability. Users can request the assistant to process tasks based on the current screen content without needing to take screenshots or switch apps. For example, in the demo video, a user opens the camera to view an indoor environment and asks for a cabinet recommendation under 1.2 meters and less than 1000 dollars, and Doubao provides a suitable suggestion based on the image and requirements.
This contextual understanding of the screen has been demonstrated by other tech companies, indicating that phone AI assistants are evolving from simply answering questions to understanding the user's specific situation and completing tasks accordingly.
In addition to interaction, personal context and memory are crucial for meeting user needs. The assistant offers a search tool for local data, allowing users to retrieve information from albums, text messages, and notes. For instance, it can determine the expiration date of a travel permit from the phone's records.
For office scenarios, the consumer version upgrades recording features and integrates with Feishu Miaojing. Users can directly query specific content from past recordings, such as device recommendations from interviews. Moreover, users can manually save screen content via voice input, three-finger swipes, or holding specific keys, creating searchable local memories.
Task execution is another core capability. For travel, users can ask for a ride to a specific location, and the assistant will confirm addresses and book the trip.
The "operate the phone" feature from the preview version is now available as a beta. Doubao notes that graphical interface agents are rapidly evolving and aims to build an open, standardized ecosystem. To achieve this, the Screen Automation Execution Protocol (SAEP) was introduced, allowing third-party apps to declare their interaction boundaries and whether they permit AI operations. Doubao respects developer wishes and will not automate operations on apps that reject them.
Regarding privacy and security, the assistant adheres to "reasonable necessity" and "user control" principles, with a layered defense system for operation scopes. A full whitepaper is available on the Doubao website.
The first consumer version of the assistant will debut on the Nubia NaviX Ultra, with pre-orders starting now and a release on September 16.