Agent Builder

1. Select your organization and the database you want the agent to use, then open Agent Builder.
2. Open the agent picker and choose Create New Agent. Enter its details and select Create Agent.
3. Add nodes to the canvas, connect the workflow, and select each node to configure its behavior. Choose models and tools that are available to your organization.

Workflow edits belong to the draft. Review save failures before leaving; running the agent waits for pending workflow changes to save.

Use the node panel to set prompts, models, tool behavior, and connections. Build a small path first, then add branches and approval decisions. Check that each path reaches the intended output and that its tools use the correct database.

Use Agent actions → View JSON to inspect the flow definition, or Edit agent to update its details. Tools remain subject to the caller’s permissions; a workflow does not grant additional database access.

1. Choose Run agent to open the chat panel.
2. Try a representative request, then a request with missing or ambiguous data. Review the answer and tool results.
3. Exercise approval and denial paths for actions that require a decision. Change the workflow and rerun the affected cases before publishing.

A successful sample conversation does not verify every input or permission combination. Test the flows your application will actually use.

1. Select Publish and enter a model identifier.
2. Review the identifier before confirming: publishing replaces an existing published agent with that identifier in the organization.
3. Call POST /v1/chat/completions with the published identifier as model, the target database ID, and authorized credentials. Follow the AI API reference for request and streaming examples.

Further draft edits need to be published before callers use them. Open Agent actions → Published agents to inspect or remove a published entry; consider applications using its identifier before removing it.