Very happy with everything-claude-code!

[jabbor]

Currently I'm using everything-claude-code with a local LLM (Omnicoder-9B Q8) running with llama.cpp, and I must say I'm impressed! This works very well! The model stays on track, checks its work, fixes the code without you having to ask it, so I'm now getting results much faster. In one afternoon the system created a python-based SQL Server MCP without major problems. The only problem I have is that LMStudio is not working well with MCP's, so it's now creating a web-ui for a MCP client. So far, I'm quite happy! 👍

[affaan-m]

This is a good use case, and it is useful to hear ECC works with a local model setup. For local LLMs, the main thing is to keep the loaded surface smaller than the default power-user install.

I would use the low-context profile, keep MCPs sparse, and disable SessionStart context if the model starts drifting:

• ./install.sh --profile minimal --target claude
• ECC_SESSION_START_CONTEXT=off while testing
• enable only the rules/skills that help your current project

Local models can do well with ECC when the instructions are narrow and the verification loop is concrete. They usually struggle when the session gets stuffed with every skill, every MCP tool, and long historical context.