Here’s what usually blocks an agent from being published and how to fix it:
Make sure the diagram is compiled• Click the blue **Run** button on the canvas – this compiles every block and will surface any errors.
Train the NLU model first• In the Prototype panel click **Train Assistant**.• Wait until the spinner beside the button finishes and you see the “trained” state.• If training fails, open the Prototype console to see which intent, slot or prompt is causing the error and correct it before retrying.
Publish only after training succeeds• Once Run → Train complete with no errors, hit the **Publish** button (top-right). This pushes the production version that the live runtime/API will use.
Extra checks • Make sure every variable has an initial value and every block has a valid next path. • Look for unused or orphaned blocks that might still reference deleted variables/intents. • If you recently added a large Knowledge Base file, remove it temporarily – an unsupported file type or very large PDF can stop training. • Refresh the tab or log out/in to clear a stuck training job.
Follow the exact order Run → Train Assistant → Publish and your agent should deploy without the “training” error.