@UmutTuzmen@ClaudeDevs Interesting point. Honest question: can you prove the default is slower than before?
Nerving the model before a new comes out seems like an industry standard, but I’ve never heard of this before.
@tickswiz@lucas_montano See a lot of ppl complaining the last weeks. If Opus now is just following user inputs and not filling the gaps, most of them may be non-technical vibe coders or just to lazy to think. Tried CC after some pause with --system-prompt-file flag today and works great with Opus 4.7
@cyb3rops@daniel_mac8@openclaw I also subscribed to Ollama Pro. Found the infrastructure is often very slow. I used it for nanobot (will upgrade to Hermes) and Open WebUi. At least here in Germany it’s not so reliable sometimes.
@ModernGrindTech@SullyOmarr I also think this is what most people do wrong. If you don’t have your OpenClaw do 24/7 things for you, I cannot think of a normal usage to reach limits at that pace.
@ModernGrindTech@SullyOmarr I keep my CLAUDE.md extremely concise - about 30 to 70 lines. Only things the agent cannot figure out on its own by reading files. Keep stuff that can outdate out.
I also use skills and MCP a lot.
But 600+ skills is insane!
What is the memory file you mentioned?