Skip to main content
The setup command writes model preference and endpoint configuration through Arka’s protected .env handling. It does not download weights or store API secrets; it prints launch commands and provider-specific next steps. Use arka integration setup <provider> for API keys, then arka model doctor to check runtime readiness. Combine local and hosted models with arka hybrid status and arka hybrid run --policy parallel. For a LAN inference cluster, configure Exo as a local OpenAI-compatible host:
Exo can pool Macs, Linux workstations, and other trusted machines. Keep its endpoint on a private network; Arka classifies it as local and never sends it to hosted fallbacks under local-only. The model advisor also annotates Apple Silicon/MLX opportunities, MoE model hints, and speculative/MTP decoding hints. These are evidence labels, not guarantees: verify runtime support and active-parameter metadata before deployment.