Maximum length hit - emsi/MyManus GitHub Wiki

When you hit the hard limit of tokens for a given conversation:

Your only option is to cleverly start a new conversation.

Use the original prompt with something like this: "You are in the middle of this... (put your original prompt here, you may need to modify it slightly or chop down to mage grammatical sens). Look for artifact files in the folders ... to understand where in the process you are and resume/continue from where you've left off".

You may be more creative of course. The idea is that model will read the TODO files it crated as well as other artifacts it made during previous run and understand the current situation and the context and resume the operation from this point.

It always worked for me but your mileage may vary. Especially if there's a lot of content and you already resumed the conversation in the past.

Sometimes you may even want to do something like this on purpose to steer the process in specific direction.