rlm

So were RLMs(Recursive Language Models) a fad or are they the bomb? alexzhang13/rlm A while ago there was hype around RLMs but now no one seems to be using it. When using opencode and seeing the context climb like crazy, I always think that it should use sub agents so that it doesn't compact as much. RLMs seem like a good solution to it, but I never seen it mentioned here in discussions and about harnesses, this makes me think it was all hype rather than a actual solution. To those who don't know what RLMs are: There is a main orchestrator and sub agents. The orchestrator and sub agents interact with a Python environment, where they can store data, execute code, and invoke other agents. The goal of the orchestrator is to assemble a answer with the least amount of tokens ending up in the orchestrator context. So the orchestrator will call sub agents and their answers will end up in variables since even the invocation of sub agents happen in python. The sub agents can also invoke new sub agents; that's where the recursive part comes from.