OpenAI compatible endpoints supported right now, designed for local agents, with potential support for other providers.

also supports subagents through parallel mid turn task delegation on multiple gpu slots, fanning out on available and queuing tasks while busy