"I’m sure there exists actually good AI tooling, but I’ll be honest, if I see a project whose description involves “LLM” or “MCP” literally anywhere, my immedia

"I’m sure there exists actually good AI tooling, but I’ll be honest, if I see a project whose description involves “LLM” or “MCP” literally anywhere, my immediate assumption is that the whole thing is vibe–coded garbage. And frankly, so far, that impulse has been correct." And yet a good, fast, problem solving local CLI llm interface is missing. Either they're proprietary (claude, codex, gemini-cli), or they're just bad, or missing (AWS ...) or both. Ollama is better than even Claude imho for just text processing but doesn't seem to have anything that can actually act on a system. Writing a bash script to do some ML task over 100 textfiles is ... pretty damn hard.