Hardware Trajectory
Android is the proof. Hardware is the direction.
Edge AI claims age badly. We label precisely what is available right now, what is in beta, and what represents our long-term trajectory.
- 1 class available
- 1 class in beta
- 4 classes planned
01Device classes
From mobile to IoT
Compatibility is decided from the device-reported profile at deployment time, not manually guessed.
Mobile Hardware (Android)
- Target Profile
- arm64, Android 10+, 4+ GB RAM
- Intended Runtimes
- GGUF (CPU), TFLite (CPU / NNAPI)
Our current product proof. A full reference agent with QR pairing and live telemetry.
IoT Hardware
- Target Profile
- arm64 Linux (e.g. Pi OS 64-bit)
- Intended Runtimes
- GGUF (CPU), TFLite (CPU)
Headless daemons for connected endpoints. Protocol is stable, packaging is iterating.
Edge AI Devices
- Target Profile
- Connected devices with local AI capability
- Intended Runtimes
- Runtime support under evaluation
Local AI support for compatible Edge AI environments. Currently in the architecture phase.
IoT Hardware
- Target Profile
- Connected IoT devices and gateways
- Intended Runtimes
- Compatible local model formats
Private, local AI for connected environments.
Lightweight IoT Devices
- Target Profile
- Resource-conscious IoT endpoints
- Intended Runtimes
- Lightweight local runtimes
Offline AI designed for practical, connected devices.
Partner Devices
- Target Profile
- Vendor-specific architectures
- Intended Runtimes
- TBD per platform
Native integrations shipped out of the box with our hardware partners.
02How we label
Three words, used strictly
Available
Shipping, supported and safe to build on. Breaking changes are versioned.
Beta
Usable, but the agent, packaging or profile may change without a migration path.
Planned
Design or research only. No agent exists you can install.
Have hardware we should support?
If you ship devices and want a LokiAI agent for them, tell us the platform, the runtime constraints and the volume.