AI Agents
Last updated
Last updated
In our ongoing effort to enhance user experience, we have introduced AI agents to our platform. Here's a brief overview of our primary agents:
Agent Allocator: This central agent, powered by GPT-3.5-Turbo, distributes tasks among three main sub-agents and their own respective subordinates. It also presents users with all the requested prompts and data.
Agent 1 - Text-to-Image (TTI): Upon receiving instructions from the allocator, TTI communicates with DALL-E to generate the requested images.
Agent 2 - Text-to-Component (TTC): Leveraging GPT-4, TTC follows the allocator's prompt to replicate or generate a model based on the given text prompt and generated image. The tasks from TTC are then divided among multiple custom sub-agents to create the requested design. The working time ranges from 30 to 260 seconds, depending on the complexity of the task.
Agent 3 - Web-Search: Based on the allocated task, this agent uses the Google Search API to gather all relevant information, URLs, and websites related to the task. These data points are then processed by an AI system, driven by GPT-3.5-Turbo, to select the most pertinent information. The filtered results, along with a summary, are then sent back to the user.
Users also have the option to choose which agent they prefer to use or to let the allocator select the most suitable agent based on the task.
Our agent allocator now stores results and history from all the other agents. However, its memory span remains limited, meaning it may erase part or all of its memory after multiple tasks (or > 8000 tokens) to make room for new information.
