Clai uses the conversation system and 'custom tools' when approaching coding tasks. This allows the LLMs to query the file system so that it builds the context specifically for the prompted usecase. Codex and Copilot etc usually takes entire repos as context, causing millions and millions of tokens, which of course is expensive. Clai, on the other hand, achieves the same result at lower cost, especially since it's possible to tweak which model to use and design specialized prompts in profiles to achieve certain tasks.

I wrote the majority of this tool over a year ago now, only now I've seen the big corporations 'catch up' with codex cli and claude sdk. So the benefit with clai here is that it's vendor agnostic.