Raw LLM Responses

Yeah, watchdog is working quite well. I always say code over prompt if possible — prompt failure rate is just too high, right. Watchdog can become much more sophisticated. My architecture is built so everything can keep scaling. New features are just another handler or plugin. Technically unlimited scale — CPU is your only bottleneck. Unknowns: 1000 agents all calling the same tools, would love to see it. Would be interesting. Onboarding agent — I'm building it. Final testing is me just using it in a container and tweaking as I go. We use Rich CLI, and have a CLI/Rich format agent — he has good knowledge on CLIs so we'll make a good effort to make it aesthetically pleasing and useful. I'll walk you through the install process, raising issues, testing, then on to creating your first project and agent. If there are any issues once onboarding is complete, you'll be given a choice to start with your model of choice — say Claude, a terminal will pop up, a Claude Code terminal, and it'll be injected with your current state. If you have any outstandings, you just chat and work through and finalize install and complete any remaining tests. Ideally you just get 100% success. I work on Linux. Mac and Windows have some testing but do need more.