A. No, since the technical configuration of the Agent in ODC Studio requires the selection of a single model (“AI Model”) to serve as the main reasoning engine B. Yes, but only through logical orchestration, where the agent uses its tools (Server Actions) to invoke other AI models during its workflow C. No, because the agent’s memory and state persistence system cannot share context between models from different providers D. Yes, Single-Agents have a native feature that allows them to automatically switch between language models to optimize the cost of each response
Yes, a single-agent in OutSystems Developer Cloud (ODC) can use multiple AI models or language models for complex tasks. The ODC platform allows AI agents to orchestrate various elements, including multiple AI models, external tools, and application logic, to achieve a larger goal efficiently. This setup is particularly well-suited for handling complex, dynamic workflows.
AI agents in ODC use the Reason+Act (ReAct) framework, enabling them to dynamically decide which actions to take based on their objectives. Within this framework, agents can invoke different AI models as needed, combining their capabilities to perform multi-step tasks effectively. For example, an agent could use one model for summarization, another for sentiment analysis, and a third for text generation, depending on the task's requirements.
The integration of multiple AI models is facilitated by using the agent-specific logic actions such as Call server action, allowing agents to pull results from various configured AI models. Developers can define these calls within the AgentTask logic to structure workflows and ensure the agent fetches the necessary context or grounding data for each model Agentic apps in ODC.
Further, ODC supports connecting AI models from providers such as Azure OpenAI, Amazon Bedrock, or custom APIs, enabling broad compatibility with different AI systems. This gives developers the flexibility to integrate multiple models seamlessly into a single-agent workflow AI models and search services in ODC.
To manage such integrations effectively, key practices such as grounding, system prompts, and state persistence help refine the agent's use of multiple AI models. These practices ensure accuracy, maintain task relevance, and enable multi-step interactions without losing context Agentic apps in ODC.
Hi André,
I think the answer is B, and the application of this is the Supervisor Agent Pattern.
Hi Duc,
The Supervisor Agent pattern is a multi-agent pattern, meaning you will have multiple agents involved. (Agentic Patterns).Considering that by "Agent" the question refers to the Agent Element added to the AgentFlow action, in fact you can provide the agent with actions that call different models to execute specific parts of the task, and the Agent can decide if calling such actions is necessary to complete the task. So, I think, option B is the correct one. The "logical orchestration" would refer to how you prepare the messages so that the model know if calling an action is necessary to complete the task.Particularly I don't like this approach if the task is REALLY complex, and in that case I would prefer to use a multi-agent pattern (like the one you mentioned) or to use a workflow to control the process. Actually, the recommendation is to not use a "Single Agent" to tackle complex tasks.But it's possible.Cheers!
On the other hand...It depends on what "...use more than one AI model..." means.The reason is that the Agent itself can "use" only one Model. It can use Tools, yes, so, indirectly, yes, it could "use" multiple models, but I think this is "stretching" the possibilities a bit too much.And for certifications, I tend to "keep it simple". If a certain question requires too much "assumptions" in order for a specific answer to be correct and that answer goes into the realm of unusual scenarios, I tend to not think of it as the correct answer.So, if this came from the certification, I would chose option A (in my point of view).
As Duc said, correct Answer is B. A Single-Agent in ODC must be configured with one primary AI Model that acts as its main reasoning engine.