Core interactions are designed to stay inside the app instead of being sent to a cloud model.
Tiny by Helix AI · Coming soon
Your AI.Your device. Your control.
Compact AI built for mobile and desktop products. Tiny is designed to work locally, keep everyday interactions private, and remain useful when the internet is unavailable.
Once included in an app, Tiny is designed to respond without depending on a data connection.
A focused Q4 model for mobile use, with voice available separately when the experience needs it.
Choose your edition
One Tiny engine.
Two ways to ship.
Use the compact edition inside a mobile app, or give users a complete local AI experience on desktop with private voice included.
Tiny Mobile
A compact local AI engine for Android and iOS products.
- Android and iOS support
- Compact Q4 model package
- Works without a data connection
- Voice available as a separate package
Tiny Desktop
A private local assistant for Windows, macOS and Linux.
- Windows, macOS and Linux support
- Offline speech input and spoken replies included
- CPU first runtime with hardware acceleration where available
- Local model, local memory and no required cloud account
Inside Tiny
Small footprint.
Serious intelligence.
A Helix tuned, 4 bit model built to bring useful AI directly into mobile and desktop products without turning every interaction into a cloud request.
Model package target
<500 MB
A compact 300 to 500 MB class AI engine for apps.
0.6BHelix tuned parameters
4 bitQ4_K_M mobile quantisation
4KContext window target
≤1.5 GBPeak memory target
Mobile + DesktopAndroid, iOS, Windows, macOS and Linux targets
Zero cloud callsLocal inference by design
Neural Edge, distilled for the device.
Tiny carries the Helix constructor pattern inside one local model: it understands the task, keeps simple work direct, checks the result, and returns one polished answer.
AssessDeconstructExecuteVerifyReconstruct
Voice built into Tiny Desktop.
Desktop included
Tiny Desktop includes local speech input and spoken replies. Tiny Mobile keeps voice as a separate package so apps ship only what they need.
Early access engineering targets. Final package size, memory and speed will be published after physical device qualification.
The Tiny advantage
Not rented intelligence.
AI you control.
Frontier APIs win on raw scale. Operating system models win on convenience. Tiny is being built for teams that want to own the AI inside their product, its version, behavior, privacy boundary and cost.
No per request AI bill.
Inference runs on the customer’s device. Growth does not turn every new user action into another cloud model charge.
Your model does not change overnight.
Ship a tested, signed model version with your app and decide when to upgrade, rather than inheriting a silent API or operating system model change.
Private by architecture.
The Tiny SDK is designed without a network client or telemetry. Local tools are denied by default and explicitly registered by the host app.
More than raw text generation.
Helix Neural Edge checks requested formats and completeness, then reconstructs one clean result instead of exposing internal model work.
What product teams need
Tiny by Helix AI
Cloud AI API
OS provided AI
Works without internet
YesNoUsually
Per request provider cost
NoneUsage basedNone
App controls model version
YesProvider controlledOS controlled
Cross platform product target
Mobile and desktopAny connected devicePlatform specific
Offline voice option
Included on desktopProvider dependentPlatform dependent
Tiny is designed for bounded tasks inside apps. It is not a substitute for frontier models where live knowledge or deep server scale reasoning is required.
Early access
Bring private AI into your product.
Tell us where Tiny could help. We will contact you when early access becomes available.
Interest registered.
Thank you. We will contact you when Tiny by Helix AI is ready for early access.