Introducing TaskWeaver
An implementation with Azure OpenAI
TaskWeaver is a code-first agent framework that uses large language models (LLMs) and generative AI to plan and execute tasks.
“Code-first” means that users‘s query will be interpreted and executed by large language models (LLMs) via code snippets that call various plugins (functions) to perform the tasks.
TaskWeaver supports rich data structures, customized algorithms, domain-specific knowledge, stateful execution, and code verification. It also allows users to create their own plugins and orchestrate multiple AI agents for complex tasks. It is an open-source project developed by Microsoft Research and you can read the original paper here.
The document introduces four key components of TaskWeaver: the Planner, the Code Interpreter (CI), the Code Generator (CG), and the Code Executor (CE). Here is a brief explanation of each component:
- Planner: This is the component that interacts with the user and manages the task execution process. It converts the user’s request into a plan consisting of sub-tasks, and communicates with the CI to generate and execute code for each sub-task…