I came across a post by He Ge titled “The Era of Large Models Has Arrived! AI Agent Enters the 2.0 Era..” In the article, Kimi’s capabilities in intelligence mining were mentioned. Recently, I tested the crewAI multi-agent system to complete an SY report for a specific ID. At that time, I only completed part of it, using two agents: the SY agent and the report writing agent. In the future, I might consider using Kimi’s capabilities for iterative optimization in this area.
The crewAI multi-agent framework allows for the configuration of multiple agents, breaking down complex tasks into smaller sub-tasks, which are then assigned to corresponding agents for implementation. Finally, using Crew to configure the workflow of the entire task will enable the agents to automate certain repetitive tasks. Each agent can utilize existing tools in the crew_tools module or develop custom auxiliary tools.
Using crewAI can facilitate simple SY tasks, such as conducting SY on a specific ID and generating the corresponding report. There are two agents: the SY Agent and the report writing Agent. Each agent has its own tasks: the SY Agent primarily searches for various historical web records related to the ID based on social ID associations and summarizes the information. It may use various keyword combinations mentioned in the prompt alongside the ID for searching. The report writing Agent analyzes the summarized search results and extracts outputs that meet the report format.
The simple demo used for testing still has many areas that need improvement, such as breaking down the agents into more detailed sub-task agents and refining the task requirements related to the report. Previously, testing with Wenxin 4.0 to complete the above tasks was generally problem-free. After the crewAI upgrade, there were iterations in model imports, and there may be some issues with the use of certain domestic models.