Google Agent Platform¶
Agent Management discovers reasoning engines in Google Agent Platform, including agents identified as using ADK or LangChain.
Connection and discovery¶
Use a Vertex AI model deployment connection with Google credentials. Configure the platform’s region and project or organization scope. The identity needs access to the reasoning engines and, when discovering projects through an organization, the corresponding resource metadata.
Telemetry requires access to Cloud Trace and Cloud Logging. Separate operational metrics are read from Cloud Monitoring. Access to the agent API alone is insufficient for monitoring.
Supported capabilities¶
Inventory: discovers reasoning engines and available platform metadata.
Interaction logs and traces: combines Cloud Trace spans and Cloud Logging entries to reconstruct interactions.
Tokens and cost: derives model usage from collected telemetry and configured model prices.
Platform metrics: retrieves operational metrics from Cloud Monitoring.
Usage: conversation counts are supported; active-user metrics are unavailable.
Uptime tests and agent reviews: supported through reasoning-engine invocation.
Limitations and platform actions¶
Monitoring depends on the traces and log content emitted by the agent. Trace collection waits for the configured stability delay before assembling recent spans.
The platform stop action permanently deletes the reasoning engine and its child resources, including sessions. Agent Management cannot restore or re-enable the deleted resources.