To address the limitations of large language models (LLMs) by enhancing capabilities in knowledge retrieval, Internet search, workflow design, and agent collaboration, Alibaba Cloud Model Studio provides the following types of LLM applications: agent, workflow, and agent orchestration. These applications can broaden the potential uses of LLMs, allowing LLMs to handle private domain questions, obtain the latest information, follow fixed processes, and automatically plan complex projects.
Application types
If you are new to LLM applications, we recommend that you start with the agent application, which boasts an easy setup and intuitive conversational outputs. By using the agent application, you can quickly build a single-function AI application.
Once you have a grasp of the agent application and require more structured AI workflows, you can try the workflow application.
Specify format: Add review and correction nodes in the workflow to ensure that the responses generated by the agent adhere to your formatting standards.
Improve quality: Add review nodes to the workflow to check technical terms and improve translation quality.
For more complex tasks, you can use the agent orchestration application. This application can automatically break down, plan, and assign tasks, coordinating multiple agents to streamline the planning and execution of complex projects.
Type | Overview | Example scenario |
Agent | A conversational AI application that completes complex tasks through contextual dialog, autonomous decision-making, and tool calling. Its simple configuration allows for quick implementation of basic functions. For more information, see Agent application. |
|
Workflow | A procedural AI application that breaks down complex tasks into manageable subtasks to enhance workflow control. You can customize task flows on a drag-and-drop canvas. For more information, see Workflow application. |
|
Agent Orchestration | A procedural application that supports multi-agent collaboration. You can orchestrate the execution logic of multiple agents or let the agents plan and execute tasks automatically. For more information, see Agent orchestration application. |
|
Function comparison
Comparison item | Agent | Workflow | Agent orchestration |
Creation method | No code | No code | No code |
Application form | Conversational | Procedural | Procedural |
Dependency on prompts | High | Low | Low |
Support | Support | Support | |
Support (official and custom plug-ins) | Support (API node) | Support | |
Text conversion | Partial support (dependent on prompts) | Support (text conversion node and script conversion node) | Support (text conversion node and script conversion node) |
Integration with external services | Support (custom plug-ins) | Support (Function Compute node) | Support |
Branching | Partial support (dependent on prompts) | Support (logical branching and intent recognition) | Support (intent recognition) |
Automatic planning | Partial support (dependent on prompts) | Not applicable | Support (agent group) |