Choose the model and context
Connect an AI account, define the task and provide the knowledge it needs. Keep unrelated customer information outside the request.
Use AI to understand a request, find context or propose the next step. Put its output into a workflow with clear tools, limits and review points.
Start with the message and relevant context.
Classify the request or return a structured answer.
Use the output to choose the next workflow path.
Add a human approval for sensitive changes.
For example, triage a new support request without handing the whole process to a model.
Connect an AI account, define the task and provide the knowledge it needs. Keep unrelated customer information outside the request.
Use a decision for classification or routing. For an agent task, configure the output schema and limits on turns, tool calls, tokens and execution time.
Connect the result to a workflow branch. Put a human approval before a change your team needs to review, and define what happens when the answer is missing or uncertain.
Run with controlled inputs. Inspect the decision, tool results and resulting service changes before publishing the workflow.
Inspect the actions and permissions available for each service before connecting your account.
Connect a supported provider in AI > Models. Provider access, model availability and billing depend on that account. The connections directory shows the setup paths available.
Configure the tools the agent is allowed to use for its task. A model connection does not automatically grant access to every service or operation.
Yes. The workflow catalog includes AI starting points. Review the template's requirements, configure the necessary accounts and test your private draft before publishing.
Start with a request your team knows. Build the next steps around it.
Follow the quickstart