OPPO has opened enrolment for Xiaobu Next, a closed beta of a rebuilt on-device AI assistant, and published the list of phones eligible for the first wave. Thirteen devices qualify — nine OPPO Find models and four OnePlus flagships — and the pitch is an assistant that stops waiting to be asked. OPPO describes the system as running a multi-agent stack and a full-domain memory system locally on the phone, so it can watch what you do and act before you prompt it.
Xiaobu Next eligible devices
The first batch of supported models, as published by OPPO on 27 July:
OPPO
- Find X9 Ultra
- Find X9 Pro
- Find X9s Pro
- Find X9
- Find X8 Ultra
- Find X8 Pro
- Find X8s+
- Find X8s
- Find X8
OnePlus
- OnePlus 15
- OnePlus 15T
- OnePlus 13
- OnePlus 13T
Two things stand out about that list. It is flagship-only — no A-series, no Reno, no Ace — and it confirms in practice what OPPO’s corporate structure has implied for a while: OnePlus flagships are inside the same AI rollout as Find phones, on the same schedule.
Eligible owners in China enrol through Settings > Help & Feedback > Co-creation, where the recruitment notice for the programme appears. This is a China-market beta; OPPO has not said anything about the feature set reaching global ColorOS or OxygenOS builds, and given how OPPO’s AI features have historically been scoped by region, that is not a safe assumption to make.
What Xiaobu Next actually does
OPPO is framing this as a shift from an assistant that answers questions to one that anticipates them. The technical claims it is making:
- On-device multi-agent system — context understanding, task planning and scheduling run locally on the phone rather than in the cloud, which OPPO calls an industry first.
- Full-domain memory — a persistent, device-side memory of the user, which is what makes the proactive behaviour possible.
- Always-on sensing — the assistant continuously reads context rather than waking only on a trigger phrase.
- 100+ system capabilities — reading and writing notes, organising the photo gallery and documents, and similar first-party actions.
- Third-party agents — an open architecture supporting external Agent, Skill and MCP integrations, with Luckin Coffee, Evernote, Fliggy and Meituan named as launch partners.
- Preset expert agents — grouped around lifestyle, work, health, finance, hobbies and parenting.
OPPO’s chief product officer Pete Lau framed proactive intelligence as the real inflection point for AI phones, positioning the next generation of assistant as a companion rather than a command line.
The part worth reading slowly
Strip the language back and what OPPO is describing is a system that continuously senses what you are doing, builds a long-lived model of you on the phone, and is wired into more than a hundred system functions plus outside services that can order coffee and food. That is a lot of capability to hand a piece of software that acts without being asked.
To OPPO’s credit, the announcement does address this rather than skating past it. The company says habit tracking requires the user’s explicit authorisation, that processing is on-device by default with cloud models reached through what it calls a secure private compute pathway only when necessary, and that critical actions — it names email and payments — still require the user to confirm. Those are the right three commitments to make, and they are more specific than most AI assistant announcements bother to be.
What is not yet clear is how any of it is verifiable from the outside. “On-device first, cloud when necessary” is a design intention, and who decides what counts as necessary is the whole question. There is also no published detail on how long the full-domain memory retains data, whether you can inspect or selectively delete what it has stored about you, or what happens to that memory when you sign out or hand the phone on. Those are the questions that decide whether a persistent device-side profile is a genuine privacy advantage over a cloud one, and a closed beta is exactly the right moment for testers to ask them.
