Overview
Doubao Phone Assistant is the system-level phone agent from ByteDance's Doubao team. It is not a standalone app but something baked into the operating system through a handset partner, which is what lets it operate across other apps. The consumer version shipped first on the nubia NaviX Ultra, on sale from 16 September 2026 at 5,999 yuan for the 12GB and 512GB model, or 5,499 with the national subsidy.
There are three ways to wake it: the orange AI key on the side of the phone, your voice, or Doubao's earbuds. The key also carries the fingerprint sensor, so verifying it is you and summoning the assistant happen in one press. Voice runs on the Seed full-duplex model, which means you can interrupt mid-sentence and correct yourself, and it handles more than ten dialects including Hokkien, Hakka, Shanghainese and Cantonese.
Execution takes two paths. One is a GUI agent that reads the screen, identifies buttons and fields, and simulates taps and typing. The other goes through interfaces, with MCP and A2A protocols connecting directly to what third-party apps expose, which the company describes as a system-level GUI Agent architecture that does not depend on simulated tapping alone. Alongside this, the platform introduced SAEP, a screen automation execution protocol under which apps can declare whether to open their screens; after the notice period ended on 14 October 2026, apps that had not declined opened by risk tier.
Key Features
- One press on the AI key: A dedicated side button with an integrated fingerprint sensor wakes the assistant once it verifies you, with no second unlock and reachable one-handed.
- Full-duplex voice: Built on Seed's full-duplex model, it can be interrupted mid-sentence and corrected, supports more than ten dialects and several languages, and was tuned for wake-up and recognition in noisy places.
- Multi-app task completion: One instruction spans apps to send messages, book meetings, summarise group chats, forward documents, compare prices or plan a trip, and tasks can continue on the lock screen.
- Personal memory: With authorisation it records preferences you state, and a voice command or three-finger swipe saves what is on screen, which you can later find again by asking in plain language.
- GUI and interface paths: It can read the screen and simulate actions, or call what third-party apps expose through MCP and A2A, taking whichever route is available so interface-backed apps respond more reliably.
- Recording into Lark Minutes: Meeting recording connects to Lark Minutes with transcription, live translation, speaker separation and automatic summaries, and past recordings can be searched to the exact moment.
Use Cases
- Recording a meeting and turning it into minutes with action items sorted by speaker instead of listening back manually.
- Asking for a summary of a busy group chat and forwarding the related documents to a colleague in one instruction.
- Booking transport and planning a route before a trip, with the calendar updated along the way.
- Saving standing preferences into personal memory so food delivery and rides do not need re-explaining.
- Handing off a long task such as sorting screenshots into folders while the phone stays locked.
- Pointing the camera or asking about the screen so it identifies what is in front of you and jumps to the matching purchase or lookup page.
Pros
- System-level integration means it can schedule across apps without a connector installed in each one.
- The AI key combines wake and authentication into a single press, which lowers daily friction.
- Full-duplex voice allows interruption and was tuned for dialects and noisy environments.
- Two execution paths run in parallel, so apps with open interfaces can use the steadier one.
- Personal memory requires authorisation, and local search is controlled by a switch.
- Recording feeds Lark Minutes, cutting the cleanup cost after meetings.
Pricing
The assistant ships with the phone at no separate subscription. The hardware is the nubia NaviX Ultra, priced from 5,999 yuan for 12GB and 512GB, 6,499 yuan for 16GB and 512GB, and 7,499 yuan for 16GB and 1TB, with a subsidised starting price of 5,499 yuan. It is currently available on this model only, and buying the phone includes the full feature set. Third-party access opens by tier under the SAEP protocol, and the range of apps the assistant can operate is visible and adjustable in system settings.
Summary
Doubao Phone Assistant takes a route only a handset maker can take: put the agent at the system layer, obtain the deep permissions, and skip the work of installing connectors in every app. Pairing the orange key with a fingerprint sensor makes waking it nearly free, and voice that can be interrupted and understands dialects moves the phone assistant from working to usable. The SAEP protocol puts the boundary in writing, letting apps declare their own terms, and whether that arrangement is accepted across the ecosystem decides how much it can ultimately do. For anyone curious, the NaviX Ultra is currently the only way to buy the full form.
Version History
- Doubao Phone Assistant added (2026-10-29): As listed: available on the nubia NaviX Ultra only, from 5,999 yuan. The SAEP notice period closed on 14 October, and apps that had not declined open for screen operations by risk tier; flight and rail booking were confirmed as supported.
- Doubao Phone Assistant consumer version ships on the nubia NaviX Ultra (2026-09-16): The consumer version launched on the nubia NaviX Ultra from 5,999 yuan, or 5,499 with subsidy. The orange AI key integrates a fingerprint sensor for one-press wake, Seed full-duplex voice supports interruption and more than ten dialects, and a system-level GUI Agent architecture runs both screen reading and MCP or A2A interface calls. Lock-screen task continuation, personal memory and Lark Minutes recording are included. The SAEP screen automation protocol lets apps declare their terms, with the notice period ending on 14 October 2026.