After experimenting and identifying a promising configuration, you can deploy your agentic system to production. Agents use the same infrastructure as Domino Apps with configurations tailored for agentic workloads.
The same instrumentation you used during experimentation continues to work in production, automatically collecting traces from real user interactions.
Before deploying your agentic system, you need:
-
Project files containing your agent code (you can optionally deploy from an experiment with collected traces)
-
The Domino logging SDK installed in your environment to enable production monitoring.
-
Project Collaborator permissions to deploy and edit agents.
You can deploy an agent in two ways: directly from an evaluation run in the Experiment Manager or from the Apps view.
After comparing evaluation runs and identifying a configuration you want to deploy:
-
In your project, from the left navigation, click Experiments.
-
Choose the experiment with the agent configurations you’re evaluating.
-
Select the run you want to launch.
-
Click … on the right to launch the run.
A deployment wizard opens, similar to the Apps deployment flow. The wizard pre-fills the code commit from the selected run and prompts you to specify the entrypoint for your agent.
Publish and share an App has information about access controls and sharing your deployed agent.
Production traces are collected automatically using the same @add_tracing decorator or init_tracing() function you used during experimentation.
No additional instrumentation is needed. Your agent code continues capturing traces from real user interactions.
The production traces appear in the agent’s Performance tab and include the same details captured during development.
-
Monitor agentic systems: Track production performance and run evaluations
-
Experiment tracking and traces with agents: Compare more configurations before deploying