Curious how you are handling the backend orchestration for the parallel agents. I built a similar multi-session workflow recently using LangGraph and found the context window costs made the unit economics tricky. Are you sharing state via Redis or keeping the contexts isolated to manage token usage?

OpenCode has been around for a while. Was there a recent update the merits posting their main website?