Cenith

On-device AI · Privacy first

AI that runs on the device, not in someone else's cloud.

Cenith is an Australian AI and SaaS company. We design, compress and ship AI models that run directly on phones, edge hardware and AIoT devices, so they respond instantly, work offline, and keep your personal data on your device instead of in someone else's training set.

What we do Work with us
  • 2021Founded
  • Gold CoastQueensland, Australia
  • On-deviceMobile, edge and AIoT
  • Privacy-firstEvery model, by design

What we do

From model research to apps people use every day.

We work across the whole stack of on-device intelligence: choosing and training the model, making it small and fast enough for real hardware, and building the product around it.

On-device models

Vision and language models engineered to fit the memory, power and latency limits of real devices.

  • Model selection, training and fine-tuning
  • Quantisation, pruning and distillation
  • Conversion for mobile NPUs, GPUs and CPUs

Mobile AI applications

Polished iOS and mobile apps where the AI is the product, not a server call behind a loading spinner.

  • Camera and document intelligence
  • Offline-first app architecture
  • Native performance tuning

AIoT & edge

Intelligence for sensors, cameras and embedded hardware that can't depend on a constant connection.

  • Edge inference on embedded hardware
  • Sensor and vision pipelines
  • Cloud services only where they add value

Why on-device

Your data should work for you, not train someone else's AI.

Most AI today runs in the cloud. Every photo, document and question is uploaded to a provider's servers, where it may be stored, reviewed or used to improve their models. On-device AI reverses that: the model comes to your data, instead of your data going to the model.

01 · Private & secure

Your data stays yours.

Processing happens on your own device. Your photos, documents, voice and personal context are not sent to a cloud provider, so they can't be collected, retained or used to train anyone else's AI model.

  • No upload needed to use AI features
  • Nothing handed to cloud vendors for model training
  • No server-side copy of your data to be breached

02 · Fast & efficient

Instant, and always available.

With no round trip to a distant data centre, results arrive in milliseconds. Models tuned for mobile and edge chips use less power and keep working without a connection.

  • Real-time response, free of network delays
  • Works offline: on a plane, underground or in the field
  • No per-request cloud cost, so products stay affordable
Cloud AI On-device AI
Where your data goes Uploaded to the provider's servers Stays on your device
Used to train AI models Possibly, depending on the provider's terms No, it never leaves your device
Response time Depends on the network and server load Instant
Without internet Stops working Keeps working

Our principle

Every model we build starts from the user's privacy and security.

Privacy isn't a setting we add at the end. It decides where our models run, what data they touch and how our products are designed, from the first prototype onward.

Focus areas

Where on-device intelligence makes the difference.

We explore applications where speed, privacy and offline operation matter most. These are the areas guiding our research and product work.

A

Vision & documents

Camera-based understanding of pages, objects and scenes: geometry correction, enhancement and recognition, all on the phone.

B

Smart spaces & AIoT

Connected devices in homes and buildings that sense and respond locally, without streaming everything to the cloud.

C

Personal assistants

Compact language and speech models that help with everyday tasks while keeping personal context on the device.

D

Industrial edge

Inspection and monitoring at the edge, where connectivity is limited and decisions need to happen in real time.

Approach

Measured on real devices, not just benchmarks.

A model that scores well on a server GPU can still fail on a phone. We validate quality, speed and memory on the hardware our users actually carry.

  1. Research & prototype

    Evaluate candidate models against real-world samples, including the hard cases, and set clear quality targets before building.

  2. Optimise for hardware

    Compress, convert and profile on target devices until latency, memory and battery use meet the product's needs.

  3. Ship & improve

    Wrap the model in a fast, simple app with safe fallbacks, then refine it with what we learn from use.

Contact

Let's talk.

Partnerships, collaborations or on-device AI projects: we'd like to hear from you.

Postal address
Cenith Pty Ltd
PO Box 5326
Q Supercentre QLD 4218
Australia