Overview
Device and model support
The exact managed model matrix, machine-wide and per-user boundaries, and the COTS validation ladder from Android phones through laptops, desktops, workstations, and servers.
Arcane OS is in active development across Microsoft NT, Linux, and Android. Device support describes which host integrations exist; it does not impose application-owned model, context, result, or hardware admission rules.
Platform integration
| Platform | Current local-model behavior |
|---|---|
| Microsoft NT | Arcane can use the machine-wide ArcaneOllama service and expose its available models through Arcane Core. |
| Linux | Arcane can use a compatible Ollama service installed and managed through the Linux environment. Arcane does not run the Linux package manager for the user. |
| Android | The current Android host does not yet expose local Ollama capabilities. Android model support requires a separate mobile runtime integration. |
Ollama publishes current platform guidance for Microsoft NT, Linux, and supported accelerators. Google's documented Android path is LiteRT-LM, which Arcane has not yet integrated.
Current behavior and target boundary
Arcane Settings presents the models returned through the selected provider's current host contract. The Settings app does not hide returned models through an additional app-owned catalog rule. Shared SDK, host, and provider code still own the catalog and can report its current compatibility or availability limits; those upstream limits are not proof that every provider model is enumerated.
The target ordinary behavior is for applications to add no prompt, context, or response cap and for a shared owner to report any provider or catalog limit honestly. Current apps pass through the context and response supplied by the shared chat path; end-to-end continuation beyond a provider's single-request boundary remains shared SDK/provider work rather than an app-local claim.
The provider and operating system still report genuine failures such as an unavailable service, an unsupported model, insufficient runtime resources, or malformed input. Arcane surfaces that result honestly and leaves model choice with the user; it does not turn estimates into a universal product gate.
Machine-wide models and per-user choices
A source pull, service installation, model download, and user selection remain different operations:
git pullupdates the checkout only.- The platform setup path installs or updates the selected Arcane development package and any platform-owned service work.
- Arcane Settings exposes download, create, load, unload, and delete actions when the configured host and provider currently support each operation.
- Machine-wide model availability is shared by that service, while the chosen provider and default model remain user settings.
Arcane does not infer that a source update changed an already installed host or service. Update the selected development package when the corresponding host code changes.
For model definitions and service behavior, continue with Arcane Ollama models. For the broader platform status, read Capabilities and current status.