Most exciting feature is that Clai may now call itself as a tool! This allows Clai to solve very dynamic problems, considering the vast set of features it has. So all tools Clai has, will become available as a tool to the LLMs. Example: we write a manager clai agent to spawn worker agents which have their own prompts and models. The manager will then on the workers completion, without tainting its own context, and may both validate and/or proceed with the results. I'll continue to expand clai in this direction, it's very interesting.

Other additions is support for more vendors, namely grok and gemini. I also added support for piping in b64 encoded images into omni models + photo models for cli-based LLM photo editing. All of this is supported recursively as well, of course, so that clai can dynamically spawn workers which generates videos and photos in omni models.

The current biggest limitation to this is the LLM models: they don't know how to use clai as a tool very well. I'll continue to expand on better documentation and instructions, these will be helpful for "normal" intelligence as well.