@matvelloso@dkarlovi@anooj Did you notice any difference in model performance (esp GLM5.2) on opencode vs provider/openrouter API? Have been hearing some chatter that the opencode hosted open models are underpowered
@NousResearch@matvelloso GLM5.2 is great but has weird bugs in the Hermes harness where it keeps failing internal tools calls - crons, memory. Switched back to Kimi K2.7 -code.
@coyington@steipete I wonder if it’s a soul.md or user.md issue. I haven’t had this issue with Codex. Gpt- 4.5 is a different issue.
On a separate note, Codex Subagents via Openclaw are a lowkey superpower.
@lucasalpes@steipete Custom subagents with unique profiles, context and steering exist in claude and codex (I think). How would what you are proposing be different?
Also why persistence- if orchestrator has to call on them. Also context rot!
Subagents in OpenClaw are lowkey its super power.
@_simonsmith I'm a long-time @claudeai user for personal uses. I had two conversations with Opus4.6 today and my usage limit capped out. Same thing happened last night. This is the first time something like this is happening to me. It is an absolute headache. Anyone know what's going on?
Really enjoyed this podcast featuring @Swats26 who talks about her journey starting @Cashkarocom@Earn_Karo . Definitely one of the best ones featuring Indian entrepreneurs.
https://t.co/uR7LMnXdr3
In all fairness Brits probably don't do Thanksgiving because if we had to take a day off for the anniversary of us fucking over a native people and ruining their country we'd never have to go to work.