Doubao Phone Assistant Consumer Version Launches, First Device to Go on Sale Sept. 16
Doubao Phone Assistant's consumer version was released on Sept. 14. It adds an AI button, screen Q&A, local memory and task execution, with 'operate phone' in beta. The first device, Nubia NaviX Ultra, will go on sale Sept. 16.
According to Leiphone, the consumer version builds on the Doubao App and works with phone manufacturers at the system level. An official video said that compared with the technology preview launched late last year, the consumer version places more emphasis on stability and daily usability. The assistant supports voice, an AI button and other wake-up methods. The dedicated AI button includes fingerprint authentication, allowing the assistant to perform tasks after the user is verified without a second unlock. Voice wake-up was also optimized for far-field, internal-noise and noisy environments.
Screen Q&A is presented as a basic capability. Users do not need to take an extra screenshot or switch apps; the assistant can use the current screen as context to continue a task. In an official video demonstration, a user pointed a camera at an indoor space and asked for a cabinet under 1.2 meters and priced no more than 1,000 yuan. Doubao combined the image and the user's requirements and found matching product recommendations on an e-commerce platform. Leiphone noted that similar screen-context understanding had appeared in Apple demonstrations for Siri and Apple Intelligence. The direction, it said, is for phone AI assistants to move from simply answering questions to understanding the user's specific situation and needs and then completing tasks.
Personal Context and memory are another focus. Leiphone reported that Doubao Phone Assistant provides a search tool for local data, allowing users to retrieve information from photos, text messages and notes when needed. For example, when a user asks about the expiration date of a Hong Kong and Macau travel permit, the assistant can find the answer from information on the phone. For office use, the consumer version upgrades the recording experience and connects to Feishu Miaojie. With local search tools, users can ask follow-up questions such as what product suggestions were mentioned in yesterday's user interview. Users can also save current screen content through voice input, a three-finger swipe up, or by pressing the AI button and volume key at the same time, creating locally searchable memories.
Task execution is the other core capability shown. In a travel scenario, a user can say, 'Call a car to my company,' and Doubao will confirm the usual address, choose a vehicle type and start the order, according to Leiphone. The 'operate phone' feature shown in the technology preview is being opened in Beta. Doubao said graphical interface agents remain a fast-developing frontier and that it hopes to build an open, regulated phone AI application ecosystem with partners. To that end, it introduced the Screen Automation Execution Protocol, or SAEP, and began a 30-day public notice period for the rules. Under the protocol, third-party apps can declare their own operating boundaries and explicitly allow or refuse screen automation by the AI assistant. Doubao Phone Assistant said it will respect developers' wishes and will not perform automated operations in third-party apps that explicitly refuse them.
The company also highlighted privacy and security. Leiphone reported that Doubao Phone Assistant follows principles of reasonable necessity and user autonomous control in data collection and processing. It has established an Agent protection system that manages the scope of operations through layers and levels. The release video said users can view the full privacy and security white paper on the Doubao Phone Assistant official website.
Ifanr, which received the consumer version for hands-on testing, reported that long-pressing the AI button wakes Doubao, while double-pressing enters real-time video dialogue. Even on the lock screen, users can ask general questions such as the weather. Sensitive operations can be combined with fingerprint verification. Ifanr said the button is not just a shortcut; it can start an action, with the current screen becoming context for the model. In tests, the assistant identified a dog breed and gave advice for beginners, summarized a paper displayed on screen, identified actor Tang Guoqiang and the character Zhuge Liang, and recommended rust-proof, minimalist bathroom shelves based on a wall's 30-centimeter gap. Ifanr said daily life and Feishu office scenarios were generally supported, while payment, privacy and irreversible operations required the user to take over and confirm. It also tested multiple tasks at once, including switching the phone to silent, setting a 3 p.m. alarm, checking the next three days' weather and giving a reminder before 7 a.m.; the assistant completed the four items in about one minute. A more complex WAIC task took 17 minutes, during which the user did not need to manage it, ifanr reported.
Ifanr also described memory as covering two types of information: data already on the phone, such as recordings, contacts, photos, messages, notes and calendar entries after authorization, and content the user actively saves, such as addresses, order numbers, travel guides and product information. Users can save content by voice command, pressing the AI button and volume key together, a three-finger swipe up, or tapping 'memory' after a screenshot, and retrieve it by long-pressing the AI button, using voice wake-up or asking in the Doubao App. Ifanr reported that users can switch to Doubao Professional's 'work task' mode for complex office scenarios and receive Word, Excel or PPT files after describing a goal. It noted that 'expert mode' and 'work task' are independent task modes; while in either mode, 'operate phone' and other assistant tools cannot be used at the same time, and users must exit the mode to call them. The memory feature is based on user knowledge and active authorization, and users can view, modify and delete saved memories, ifanr said.
Ifanr characterized the direction as smartphones becoming 'action terminals,' with Skill becoming a new kind of App and the system managing background execution, task queues, a task center and fingerprint confirmation. It said the technology preview showed the imagination of AI operating phones, while the consumer version must answer whether ordinary users are willing to use it every day. The first device, Nubia NaviX Ultra, is already available for reservation and will go on sale on Sept. 16, according to Leiphone.