Platforms¶

A platform is a configured connection to an agent provider in Agent Management. It provides the credentials and provider-specific configuration used to discover agents and collect telemetry.

Creating a platform¶

From Platforms, create a platform and configure:

Platform type, which identifies the connected agent provider.
Platform display name, which identifies the connection in Agent Management.
Connection name, which provides the credentials used to access the platform.
Data Storage project, which stores backing datasets and Flows for agents on the platform. Agent Management can create this project or use an existing project.

Creating the Data Storage project automatically requires project creation permission. An OpenTelemetry Receiver platform does not require a platform connection; it receives telemetry from instrumented agents. See OpenTelemetry Receiver.

Supported platform types include Dataiku, AWS Bedrock, Microsoft Foundry, Microsoft Copilot Studio, Salesforce Agentforce, Databricks, Snowflake Cortex AI, Google Agent Platform, and OpenTelemetry receivers. The supported discovery and telemetry capabilities vary by platform and product version.

Scanning and discovery¶

Agent scans connect to the platform and update the inventory with the agents currently available to the connection. Scans can run on a schedule or on demand with Scan now.

The platform page reports whether scanning is enabled, the result of the latest scan, and the next scheduled scan. A scan can discover new agents and update platform metadata. It does not silently delete existing inventory records when an agent is no longer returned by the platform.

Platform activity¶

The Activity tab records background operations for the platform. The principal operation types are:

Agent scan: Lists agents on the platform and reconciles them with the inventory.
Log fetch: Retrieves agent interaction logs and stores them in backing Dataiku datasets. Fetches can be incremental or cover a historical range.
Operational metrics fetch: Retrieves platform metrics such as latency, error rate, token volume, and interaction volume when the platform exposes them.
Uptime test: Calls an agent periodically and aggregates the result into an uptime metric.

Each operation records its start and end time, trigger, status, and error details. Collection failures do not necessarily mean that the agent itself is unavailable. Review operation failures separately from metric values and alert occurrences.

Platform logs¶

The platform Logs page provides the detailed internal log files behind the operation history. Use it when an Activity entry identifies a failed operation but does not provide enough detail to diagnose the cause. Logs can be filtered and downloaded individually or as an archive.

Platform settings¶

Platform settings control the platform connection, scan schedule, data collection, uptime tests, platform alert configurations, and default permissions. Agent-level settings can further specialize supported collection and monitoring behavior.

Uptime tests¶

For platforms that support uptime tests, configure the test question, timeout, and schedule on the platform. Enable testing for the agents to include. Each test actively queries the agent, so its result measures the configured interaction rather than every user request.