Ming Zhou@rrfe

It supports using plugins to call external resources (such as internet search and knowledge base retrieval) and modifying the results returned (generating images and code, etc.) The powerful AI capabilities are used as the core CPU of the system for data analysis and task processing. Before sending messages to the AI, interception can be done to add additional information such as relevant knowledge base information (such as hwchase17/langchain), network search information, and even adjusting the request data to support multi-modal recognition. For example, image information can be parsed by other tools and analyzed by Claude after obtaining the corresponding information. After the AI responds to the question, the response result can also be processed again, such as using Stable Diffusion for image drawing or using Gmail's API for email sending. To enhance the flexibility of the implementation, Hooks are used to achieve corresponding capabilities. Before and after the user's information is handed over to Claude for processing, chat information is passed to plugins for processing. The pre-processor can modify the user's question to add context information or even directly modify the question. After that, the pre-processor submits the processed information to Claude in a specific format. Once Claude returns the information, the post-processor can analyze and process the response or call APIs from other platforms to process tasks. In this case, we need to adjust the Prompt in the pre-processor so that Claude returns Action-formatted data. { "type": "createMail", "payload": { "content": "xxx", "target": "[email protected]", "platform": "google" } }