Dense parameters
Qwen 3.8 converted for local RKLLM inference.
INDEPENDENT EDGE AI PROJECT
Fully offline AI.
Running on real edge hardware.
ASLIX™ is a self-contained local AI platform combining voice, memory, geographic intelligence, vision and on-device inference without depending on cloud services.
MODELQwen 3.8 27B Dense
DEVICENanoPi M6 / RK3588S
RUNTIMERKLLM / 3 NPU cores
NETWORKOffline capable
THE RESULT
A 27B dense model running locally on RK3588S.
This is not a cloud demo. The model is loaded and inferenced directly on the edge device.
PHYSICAL DEVICE
ASLIX running on the actual hardware.
This project is tested on the physical NanoPi M6 platform, not simulated in a cloud environment.
Quantization
Running through RKLLM Toolkit / Runtime 1.3.0.
NPU cores
All RK3588 NPU cores active for inference.
System memory
ASLIX and the local model sharing one compact device.
SUSTAINED GENERATION~1.07–1.10 tok/s
PREFILL~10.38 tok/s
CONTEXT2048 tokens
MODEL LOAD HEADROOM~3.6 GiB available
Measured on the physical NanoPi M6 during the current tested configuration. Numbers may change as optimization continues.
WHAT ASLIX DOES
More than a model running in a terminal.
Voice
Local speech recognition and locally synthesized spoken replies.
Memory
Persistent local conversation and user context maintained outside the language-model process.
Geographic intelligence
GPS, offline points of interest, location context and navigation-related functionality.
Vision
Local camera and vision-model workflows for image understanding when requested.
Sensor context
Local sensor information can be combined with AI responses while remaining on-device.
Resource orchestration
Models and optional workloads are coordinated around constrained edge memory and accelerator resources.
ENGINEERING
Built through testing, failure, conversion and iteration.
ASLIX is independently developed and tested on physical edge hardware. The project focuses on making capable local AI practical where privacy, connectivity and hardware constraints matter.
Public project information includes benchmark results, hardware configuration and reproducible observations. Proprietary conversion and deployment know-how is not published.
INTELLECTUAL PROPERTY
ASLIX™
ASLIX is a trademark-pending project. Patent protection is being pursued for aspects of the system architecture and operation.
Nothing on this site is intended as a dedication of unpublished proprietary implementation details to the public.
GIVE BACK
Support the communities that helped make ASLIX possible.
ASLIX grew from independent research, open-source software, documentation, technical communities and developers willing to share what they learned.
20% of community-support contributions will be set aside to support open-source projects, independent developers and technical communities that helped make ASLIX possible. The remaining 80% supports continued ASLIX development, including hardware, testing, models, hosting and research.
The support program is not active yet. Distribution details and recipients will be published before contributions are accepted.
CONTACT
Interested in ASLIX?
Engineering collaboration, project questions, licensing discussions and business inquiries are welcome.
aibender100@gmail.com