Skip to main content

Agent Instances

An agent instance is a running or completed execution of an AI agent defined in your models. As agents take part in your processes and cases, Hub records each one so you can monitor what the agents are doing, review the model calls they make, and check that any guardrails and evaluators are passing.

Agent instance list​

The Agent Instances view lists every agent execution in the environment, with its Id, the Reference ID of the work it belongs to, the Definition Name of the agent that produced it, and when it was Created. Narrow the list with the Agent definition selector, the Latest chip (latest agent-definition versions only), and filters for ID, reference ID and start user; a tenant picker is available in multi-tenant environments.

The list of agent instances

Agent instance details​

Click an agent instance to open its detail view. The tabs shown adapt to the kind of agent (Hub recognises orchestrator, utility, knowledge, document, A2A and external agents, plus guardrail and evaluator "judge" agents):

  • Details — the agent definition (with a link to it), and the reference type and ID of the work it is bound to.
  • Invocations — each call the agent made, with the user, start and end times, the operation key, and the prompt and output token counts. You can download an invocation's timeline as a PDF for a full trace.
  • Evaluations — for agents that run guardrails or evaluators, the verdict of each check (Passed / Failed / Running), its score and reason. (Hidden for judge agents, which are the evaluators.)
  • Knowledge Base Links — shown when the agent references a knowledge base, listing the linked entries.

An agent instance detail view

Token usage recorded here feeds the cost figures on the Agent dashboards. Costs and model pricing are configured under Configuration → Agent Model Prices, and connections to AI providers under Configuration → Agent Connections.