On-device or local
The model runs on the same computer or device as the application, through the local gateway on 127.0.0.1. This is the path used in the developer preview.
Deployment options · Design direction
Choose where the model runs.
Model size is one choice. Where it runs is another.
Adil Edge separates the model you choose, Adil Eco, Core or Boost, from where it runs. Each deployment option below is a design direction; none is a released product or service.
Three deployment options
Device, server or private cloud.
Self-hosted server or VPS
The model runs on a Linux server that you operate and serves approved applications on your network. Remote access is off by default and would need authentication and encryption that you configure.
Private cloud
The model runs in cloud infrastructure that you control, under the same controls as a self-hosted server. This option is planned and has not been tested.
Engineering objective Important: Multi-tenant serving, concurrency guarantees, production security assurances and deployment tiers remain engineering objectives until tested.
Local inference and device requirements
What has been measured, and what has not.
No minimum system requirements have been published for any Adil model.
Adil Eco has been measured on one Apple M1 (8 GB) test host only. Its 4-bit file is about 484 MB, but on the Apple M1 the process needed more memory than the file: about 867 MB at a 4K context and about 1,331 MB at the default 8K context, because the runtime reserves a context cache that grows with the context length.
Performance on x86 laptops, including the ThinkPad T480, and on Linux and Windows hosts is unverified.
Adil Core and Adil Boost have not been trained, so no requirements exist for them.
The developer preview runs inference on the CPU; a GPU is not required by its design.
Why CPU-first matters
Dedicated accelerators are valuable, but not every useful business task requires one. Adil Edge explores how compact AI workloads can make better use of ordinary hardware, particularly where privacy, low deployment complexity and predictable infrastructure budgets are priorities.
Designed to stand alone
Every deployment option is intended to function independently of any central AI orchestration service. Future integrations may enable additional routing or governance, but they are not prerequisites for the core offline/local execution ambition.