Hello,
How is Agent Workbench Execution officially calculated for licensing purposes?
In ODC Portal > Analytics, I can see "Top agents by request". Is the Requests count equivalent to Agent Workbench Executions, or is Execution calculated differently?
If it is different, where can I view the actual Execution consumption?
Thank you.
Hi @mkmk For OutSystems ODC Agent Workbench - Execution Counting (Simple Explanation)
- 1 Agent Execution = 1 CallAgent execution at runtime.
- Every time your app calls an Agent, 1 execution is consumed.
- Even if the Agent uses multiple tools/actions internally, it is still counted as 1 execution.
- Even if the AI model is called multiple times during reasoning, it is still 1 execution.
- In ODC Analytics,"Top Agents by Request" generally shows how many times the Agent was called.
- You can view total Agent Execution consumption in ODC Portal → Management → Admin → Subscription → Agent Executions.
Example:
User asks a question → Agent called once → 1 Execution
Agent calls 5 tools and 3 AI reasoning steps internally → still 1 Execution.
Thank you for your explanation. I checked the location you mentioned and was able to find the information I was looking for. Your answer was very helpful.
Hello @mkmk,
OutSystems is essentially a development platform and a bridge between the application and external AI systems/models. In ODC, an external AI model is integrated into the application through an Agent.
The Agent acts as the ODC-level interface for the AI model. It contains things such as the system prompt/instructions, response structure, and other configurations required to define how the agent should behave.
Based on my understanding, the “Top Agents by Request” located in analytics is related to the agents we create within the application logic and represents how many times the Call Agent action is invoked. In other words, it indicates how many times our application actually calls the AI Agent.
The AI model itself is different. The Agent acts as the layer between our application and the underlying AI model. For example, we may make one request to the Agent, but internally the AI model could be invoked multiple times perhaps 5 or 6 times depending on the agent's reasoning, loops, decision-making, tool usage, or other processing.
These underlying model interactions consume tokens, and the way this consumption is calculated depends on the specific AI model/provider. For example, OpenAI has its own usage and token calculation, while Anthropic/Claude has a different usage model.
So, in simple terms, I understand it as:
Application → Agent Request → Agent → Multiple AI Model interactions → Token consumption
Therefore, Agent Requests and AI Model usage are two different measurements: the first represents how many times our application calls the Agent (Requests), while the second represents what happens behind the Agent and the resources/tokens consumed by the underlying AI model.
You can find "Top AI models by request" next to the Agent request card which indicates the AI Model calls itself.
Thank you for your explanation. It was very helpful.
Hi @mkmk,
I believe the aswers is documented here: Agent execution - OutSystems Support
Regards,
Daniel
Thank you for your explanation and for sharing the documentation. It was helpful for understanding the relationship between Agent requests and AI model requests.