Pinned
The tl;dr bullet point version of Weave-Agent is:
- You organize a ReAct agent with a second reasoning and action stage for checking the result of the first action.
- Each stage of the agent is a python code block and the whole framework is presented as a long python program
The ReAct loop on the left is how most LLM agents are implemented, it fails because the reasoning desynchronizes from the problem state. I attempt to fix this by having the agent write down its expectations for the action and check its work with unit test callbacks.
