Today, ByteDance has shattered the ceiling of mobile AI assistants: the new generation of "Doubao Mobile Assistant" technical preview has made its debut! This is not just another voice assistant, but a true "second brain for your phone," which can see, remember, and act, even helping you master your entire phone.

image.png

The real "on-device memory": it remembers life better than you

Doubao Mobile Assistant has achieved persistent Memory functionality on the device for the first time, with all memories encrypted and stored locally on the device, and can be turned off at any time with one click.

The actual results are amazing:

- Question: "Where did I park?" → It directly shows you the photo of the parking spot you took last time + floor guidance

- Question: "What is the pickup code?" → It instantly reads the delivery SMS and tells you "5872"

- Question: "What is the high-speed rail seat number?" → It automatically finds the 12306 record: "9 car 12A by the window"

- Even remembers that you "like Van Gogh," and when planning a trip to Paris next time, it will prioritize the Orsay Museum

Cross-App automation: one sentence lets the phone do things on its own

This is the most powerful feature of Doubao Mobile Assistant: AI can take over the screen like a human, performing automatic clicks, inputs, and swipes across apps.

Real demonstration cases:

- You say: "Compare prices for this hair dryer across the web" → Doubao instantly opens Taobao, JD.com, Pinduoduo, and Douyin Shop, giving the lowest price within 3 seconds and stopping at the payment page

- You say: "Help me take three days off, and also book a train ticket back to my hometown" → Automatically opens DingTalk / Feishu to fill out the leave application → Submit approval → Jump to 12306 to book a ticket → Complete payment

- Even Tesla owners were shocked: one sentence "open the front trunk to put something in" → Doubao remotely controls the vehicle to execute it

Real-time multimodal interaction: AI can "see" the camera and start talking immediately

Pick up an English picture book and point the camera at it, and Doubao Mobile Assistant immediately activates the real-time video call mode:

- The screen displays bilingual Chinese and English subtitles

- AI tells the story in fluent Mandarin or English, asking questions as it goes

- Can also adapt the plot based on the child's reaction on the spot

Pro Mode: complex long-chain tasks, all done with one command

For vague requests, Doubao directly activates "Pro Mode," combining GUI simulation clicks + API tool calls + strong reasoning to complete tasks that AI would never dare to handle before.

Ultimate case for a Paris trip:

User instruction: "Next month I'm going to Paris, mark all the restaurants I've saved on the map, and help me book a museum ticket with my favorite exhibition."

Doubao's execution process:

1. Read memory: the user loves Van Gogh

2. Search current exhibitions: the Musée d'Orsay is currently featuring a Van Gogh exhibition

3. Open Gaode / Google Maps and mark all the Michelin restaurants you've saved

4. Jump to the official website and successfully grab the ticket

5. Generate a complete itinerary and push it to the notes app

Privacy at maximum: all memory is localized, can be turned off with one click

ByteDance repeatedly emphasizes: memory data is processed and stored locally on the phone, not uploaded to the cloud, and users can completely turn off the Memory function in settings at any time, achieving true "control and trust."