I am building a full fledged RLM harness in rust using fast-rlm
It's coming along nicely. Every context, prompts, file content is a python variable. Every tool runs inside the python repl. It's WIP but super excited about it!
PS: deepseek-v4-flash is pretty good as an rlm