@rodydavis Per subagent model + effort pinning, including bring your own endpoints, would at least let me port my setup to at least give it a fair shot
I’d love to give it a go and really experiment with how good their cloud environment set up is, always wondered if their platform would be the best for a fully automated tested/development suite for someone with cross device testing, also see how good their own model is, can never beat exploring the limits of what the true best “free model” is!
I run mine as three tiers:
Daily: Local Qwen 3.6 27B scouts arXiv 24/7 -> cards + relevance score
Daily: DeepSeek V4-Flash -> reviews the shortlist. If anything seems to wild as a claim it’ll investigate the details further;
Weekly: Opus sends Haiku subagents to verify OMP configs/agent files/flows/skills then writes the report on which could benefit for these; findings.
I’ll review in case there’s anything I personally don’t want if there’s big changes, I’ll review or perform some testing
cheap at the bottom, smart at the top, tbf could probably do with switching DeepSeek flash job to weekly as a bulk review
@0xSero@VictorTaelin Great advise, if you’re not having your long running research agent stalking it for potential improvements to your entire setup you’re missing out