Agent runs

Inspect agent sessions, conversations, and traces.

An agent run is a recorded agent session. It captures the prompt conversation, tool calls, model activity, status, usage, and trace data for one session.

Use agent run history when you want to understand how an agent responded, why a tool was called, where a session failed, or what a running session is doing.

Where to find agent runs

Open History in the web app, then filter Type to Agent. Select a run to inspect its details.

The History page is shared across agent and workflow runs. See Run history for the full table, filters, statuses, and detail tabs.

What starts an agent run

Agent sessions can start from several places:

Source	Example
API or CLI prompt	keystroke agents prompt support --message "Help me"
Gateway	A Slack thread routed to an agent
Trigger	A webhook, schedule, or poll attachment that invokes an agent
Workflow	A workflow step that calls Agent.prompt()
Another agent	A subagent tool call

Follow-up prompts with the same sessionId continue the same session instead of creating a separate one.

Inspect the conversation

For agent runs, the detail panel’s Overview tab shows the recorded conversation. User and assistant messages are shown in order, and assistant messages can include collapsed details for:

Detail	What it helps with
Thought	Reasoning or thinking output recorded by the model
Tool call	Tool name, arguments, and result or error
Error	Session error payload for failed runs

The Metadata tab shows run identity, actor, project, trigger, timing, models used, services used, and the billed amount. Model usage on your own provider keys (Managed services) is labeled Your API key in the usage breakdown and excluded from the billed total. The Trace tab shows raw spans and logs when trace data is available.

Conversation history depends on what the runtime recorded for that session. If no messages were recorded, the detail panel shows an empty conversation state.

Use the CLI

Use keystroke agents sessions when you know which agent you want to inspect:

keystroke agents sessions list support --status failed
keystroke agents sessions get support <session-id> --wait --timeout 120
keystroke agents sessions get support <session-id> --summary
keystroke agents sessions get support <session-id> --include messages,trace

Use --wait for an active session instead of polling it with a shell loop. If the bounded timeout elapses, the command returns the latest running state with exit code 0; run the same command again to continue waiting.

Use keystroke history when you want cloud history across agents and projects:

keystroke history list --kind agent
keystroke history get <run-id>

agents sessions inspects sessions for a specific agent on your deployed project. history is for platform run history across the active organization. See the CLI reference for command details.

Next steps

Run history

Learn the shared History table and detail panel.

Run agents

Start and inspect agent sessions.

Build agents

Define agents, tools, models, and runtime behavior.

Workflow runs

Compare agent sessions with workflow executions.