The Microsoft AutoGen open-source framework Magentic-One CLI is designed for high-level planning, guiding other Agents, and tracking task progress. It features a layered architecture with multiple software interfaces to meet different scenario requirements.
Using Magentic-One CLI
Core Design Philosophy:
The process is as follows: The operation of Magentic-One is based on a multi-Agent architecture, where the chief coordinator (Orchestrator) Agent is responsible for high-level planning, guiding other Agents, and tracking task progress. The coordinator first develops a plan for handling tasks, collecting necessary facts and reasoned guesses in a task ledger. At each step of the plan, the coordinator creates a progress ledger, reflecting on task progress and checking if the tasks are completed. If tasks are not completed, it assigns a subtask to other Agents of Magentic-One to complete. Once the assigned Agent completes the subtask, the coordinator updates the progress ledger and continues in this manner until the task is completed. If the coordinator finds that progress is insufficient, it can update the task ledger and create a new plan. As shown in the figure, the coordinator’s work is divided into an outer loop for updating the task ledger and an inner loop for updating the progress ledger. Using in the command line terminal: First, install the dependencies with pip install -U magentic-one-cli. Directly run m1 “I am in a very good mood today” for Q&A. export OPENAI_BASE_URL=”https://yunwu.ai/v1″ export OPENAI_API_KEY=”sk-COUbOl2EH7IYaz6SqFADUp7Ie4WA4rzFvNr3PzIAZjVyHQcJ” Integrating for use in code: Relevant test code can be found in the MagenticOneCli folder under 1_BasicTest.py.