Streaming, and interruptible.
If Hojo is mid-answer and the user starts talking, it stops and listens.
Your earbuds, ready to talk back.
Interactive audio for brands that make earbuds, AI glasses and speakers. Real-time conversation, translation and voice control run on Hojo's own speech models and Agentic OS, so your team can focus on the hardware.
Like talking to a person: cut in any time.
Users speak to their earbuds and Hojo answers right away. The conversation streams in real time, can be interrupted at any point, and works in more than one language.
More than one language, understood.
Speech recognition covers Chinese and English, plus German, French, Italian, Portuguese and Spanish, and it understands Cantonese and Sichuanese.
Face to face or on a call, in two languages.
Meeting a client, travelling, taking a call in another language: translation is built into your product. It works face to face, and on calls it works both ways.
Face to face, each in their own language.
Put the phone on the table and turn the top half toward the other person. What you say becomes text as you say it, and they read it in their language.
On a call, translated both ways.
When they speak, you hear it in your language; when you speak, they hear it in theirs. With earbuds on Hojo, call translation is a switch on the device page in the app.
Say it once, and Hojo gets it done.
Users only ever talk to Hojo. Translation, meetings and scheduling go to dedicated agents, and the result becomes reminders and follow-up.
Hojo at the centre, specialist agents around it.
Agents can be added or swapped. When your product needs a new ability, a new agent plugs in.
Hears it, then does it.
Conversations turn into notes, reminders and follow-up, so what was said keeps moving.
Your earbuds get their own page in Hojo App.
Pair once and your earbuds get a device page in Hojo App under your product's name. Earbuds already sit in the call's audio path, which is what makes call recording and two-way call translation possible.
Call audio stays standard.
The call itself, over the standard Bluetooth call profile. The copy taken for Hojo sits beside it and doesn't change call quality.
The private link is checked, not assumed.
Live audio, device status and stored recordings between earbuds and app. A Bluetooth icon showing “connected” doesn't mean this link is up, so the app checks both.
Cloud AI can wait.
Transcription, translation, voices and notes. It can be slow or unavailable; it can't take the recording down with it.
Listening, speaking, remembering: all built by Hojo.
Voice comes in through Hojo ASR and becomes text. Agentic OS understands, answers, hands work to agents, and reads and writes memory. The answer goes back out through TTS-Light.
Our own speech models, second worldwide at best.
Hojo ASR's best result on the Open ASR Leaderboard was second worldwide (June 2026). TTS-Light is a light speech synthesis model that runs on a CPU.
The family also includes Chat-Duplex and Omni.See the modelsAgentic OS understands and acts.
It understands speech, keeps the context and directs specialist agents. Earbuds, glasses and phone share one context.
About Agentic OSA memory framework that keeps the background.
Conversations become long-term memory, and every entry keeps its source. The next time the user speaks, Hojo already knows what came before.
About the memory frameworkThree ways to bring Hojo to your product.
Use Hojo App as it is, build it into your own app with the SDK, or have an app made under your brand. The difference is who builds the app and whose brand it carries.
Use Hojo App.
You provide product names, renders, the capability list and protocol documents. Pairing, sync, error recovery and testing on real devices are on Hojo.
Devices with complete materials within the standard support list go live in one business day.Hojo App for hardwareBuild in with the SDK.
Keep your own app, interface, membership and distribution, and add Hojo models and Agentic OS capabilities.
For teams with mature R&D and operations that need full control.Get an app in your brand.
A branded app generated from your assets, with interface templates, business-model setup, multi-platform source code and test builds.
For brands that want an app of their own.