Google Vertex AI agent connector features table

See the following table for an overview of the capabilities offered by the connector when executing agents in agentic processes.

Feature Area	Feature	Google Vertex Agent in Maestro
Authentication	Single token, Bearer token (e.g. Personal Access Token)	No
Authentication	Service Principal-based authentication (e.g.,service account key)	Yes
Authentication	User-delegated, token-based authentication (e.g., On-Behalf-Of, OpenID).	No
Project Discovery	Discover available projects based on authentication	No (provided by the user)
Agent Discovery	Discover available agents	Yes (lists all AgentEngine agents within a project)
Invocation	Supports synchronous (request-response) invocation.	Yes
Invocation	Supports streaming invocation (Server-Sent Events).	No
Invocation	Supports asynchronous invocation with polling or webhook/callback.	No
Invocation	Ability to specify the agent's tools/skills within the API request.	Yes (in Prompt)
Session Management	Maintains conversational context via session IDs and allows for the retrieval of event history from an ongoing session.	Yes (use of session id within Managed Properties in Maestro)
Output Controls	Enforces structured, schema-adherent JSON output from agents.	No
Content Handling	Supports multi-modal content exchange (e.g., binary files, images, audio).	No
Observability	Provides comprehensive observability through togglable verbose tracing, OTEL-formatted traces, and cross-platform correlation IDs.	Yes (in Google Cloud Console)
Consumption & Attribution	Tracks and attributes AI consumption (tokens/units) per response for cost analysis.	Yes (the token cost of each response is included in the {:}response)
Versioning	Enables agent versioning for CI/CD workflows, targeted invocations by version number (e.g., GA, preview), and rollbacks.	No
Error Handling	Delivers structured, detailed error messages (reason, domain, metadata) for advanced debugging and conditional logic.	Yes