Agent Chat

Agent Chat is not an interactive chat with an AI about the process instance — it is the recorded history of every AI Agent Connector session that actually ran as part of it. Whenever a service task in the process uses the AI Agent connector (cibseven-ai-agent), each request, response and any error is appended to a per-activity chat-log variable, and the Agent Chat tab in the Process Instance View renders that timeline back to you, in order. See Audit Trail & AI Transparency for exactly what is recorded.

For each recorded activity, you can see:

• the messages exchanged (system, user and tool) for every agent invocation on that activity
• which tools were available to the agent, and which it actually called — including CIB seven processes exposed as MCP tools
• any process instance the agent started as a side effect (via ProcessStarterTool), with the resulting process instance ID and the principal it executed as

Because the timeline is built from the same chat-log variable used for compliance record-keeping, it is a faithful account of what the agent was told and what it decided during this process instance — not a live conversation you can steer. Treat AI-generated outcomes accordingly: review them against the underlying tabs of the Process Instance View before acting on anything consequential.

Use in EU AI Act traceability

For regulated processes, Agent Chat is where the traceability question — which model ran, with what prompt and parameters, and what it did — becomes visible to a human reviewer, rather than staying buried in a process variable. Combined with a human-review task before consequential effect and the process instance’s own history (assignee, business key, completion time), it gives an auditor a readable trail of what the AI did and who signed off on it. See EU AI Act — traceability & documentation for what is captured automatically versus what remains a process-design decision.