Monitoring Activity

The Activities page is a live monitor for every task running across your Solace Agent Mesh deployment. As a task moves from a user request through agents, large language model (LLM) calls, tool invocations, and peer delegations, the Activities page renders that execution in real time as a visual flow graph, an ordered list of steps, a raw event stream, or a performance timeline. Use it to watch a task as it happens, to trace a completed task end to end, and to reach the artifacts a task produced.

This page is for operators and developers who debug live tasks. For sending messages and managing conversations, see Chatting with Agents.

Opening the Activities Page

  1. On the navigation bar, select Activities. If the navigation bar shows a Signals & Usage group instead, select Signals & Usage, and then select Activities.

The page lists recent task history as sessions, newest activity first. A session is one conversation: every message you send in a chat runs as a task, or turn, of that session, and a session row expands to list its turns. A task that ran outside any session, such as a webhook run, forms a session row of its own.

The page loads the list from recent task history rather than from a live connection; click Refresh to update it. The search field finds a session, task, or user by its exact ID. The list shows a task found by its ID among the other turns of its session.

Which tasks the list shows depends on your role. You always see the tasks you submitted. For each agent on which your role grants Can view activities, you also see every top-level task that started with that agent, regardless of who ran it, including tasks that an entrypoint runs as a system user. The grant does not include the sub-tasks that the agent ran as a delegated peer of another agent. If your role holds the activity:*:read scope, you see every task. When you can see other users' tasks, Show my sessions only narrows the list to your own sessions. For more information about roles and scopes, see RBAC Reference.

With authentication turned off, as in the desktop application, every task is available, but the list starts narrowed to your own sessions. Clear Show my sessions only to see tasks that ran as a system user, such as webhook tasks.

Selecting a Task

Expand a session row and select a task to open its details beside the list: the flow graph and the step detail. Each task row shows the initial request text, an event count that climbs as the task produces events, and, after the task has reported its usage, the number of tokens it sent to and received from the model. Hover over a row to see the task ID and the time of the last update. A row also names the agent or workflow that ran the task and, unless the task is a sub-task, what started it: the channel the request arrived through, the entrypoint that admitted it, and the sender's event name, for example Dependency update · webhook · github-deps · check_suite.completed. When an agent delegates to a peer agent, the list nests the resulting sub-task under its parent.

The header of the selected task repeats what started it: the channel and the entrypoint; for a webhook task, the route that matched; and for a request from GitHub or GitLab, the sender's event name and delivery ID. If your role includes the entrypoint:*:read scope, the entrypoint name links to the entrypoint's page. For more information about entrypoints, see Configuring Entrypoints.

A task that is still running streams updates live. Selecting an older task loads its full history from the server so you can replay exactly what happened.

What you can open for a task also depends on which grant gives you access to it. For the tasks you submitted, and for every task when your role holds activity:*:read, everything on the page is available, including the sub-task rows nested under a task. For a task you see through a Can view activities grant, the list row shows its status and initial request text, without sub-task rows. Its flow graph, step detail, recorded request and response, and task log download open only when your role grants Can view activities on every agent that took part in the task, and the sub-tasks then appear inside the flow graph. Live updates of a running task and the session title are available only for the tasks you submitted and, when your role holds activity:*:read, for every task.

Your Can view activities grants must cover every agent that took part in a task before you can open it. If a single participating agent is outside your grants, the page reports that the task details are not available and does not say which agent is missing, because that agent's prompts, LLM responses, tool arguments, and tool results would otherwise be exposed.

Asking the Activity Monitor About a Task

The Activity Monitor is an experimental feature. For more information, see Chatting with the Activity Monitor (Experimental).

Ask AI in the header of the selected task opens a new chat with the Activity Monitor, with a message that names the task ID and, for a task that ran in a session, the session ID. The message carries only the IDs, never the task's content; the Activity Monitor reads the task itself and shows you only what your role permits. You can edit the message before you send it, and ask about any other task by its ID in the same chat.

Ask AI is unavailable when your role does not permit the Activity Monitor to read the task on your behalf.

The Views

The Activities page offers four ways to look at the selected task. The flow graph is the central visualization; the linear step view, the raw event stream, and the Gantt timeline are complementary detail views.

Flow Graph

The flow graph renders the task as a pan-and-zoom node graph. Each node is a participant or step in the execution (the user, an agent, an LLM call, a tool call, a peer delegation, or a workflow construct such as a switch, map, or loop), and the edges connect them in order.

  • Center Workflow re-centers and fits the graph to its panel.

  • Detail Mode toggles nested agent details on and off, so you can collapse a busy graph down to the high-level shape or expand it to every internal step.

  • Click any node to open its Node Details. Clicking a node or edge also highlights the matching step in the step detail.

For a workflow task, the graph draws every node in the workflow, including the ones that did not run:

  • A node that failed has a red outline.

  • A node the workflow skipped is dimmed with a dashed outline and labeled skipped.

  • The Node Details of a tool node show the tool it called, the arguments it passed, and the result it returned.

The workflow skips a node because its condition was false, because a decision chose another branch, or because everything the node depended on was skipped. The Node Details of a skipped node give the reason.

For tasks that ran before Agent Mesh recorded the position of skipped nodes, the graph places each skipped node using the workflow's current definition, or the skip reason when the definition is not available. If the workflow has changed since the task ran, such a node can appear in a different position than it had when the task ran, and its Node Details say where the position came from.

Step-by-Step Linear View

The step detail lists the task's steps top to bottom under the Events heading, each with a millisecond-precision timestamp. The card for each step shows detail appropriate to its kind:

  • The original user request and the agent's final response, rendered as formatted text.

  • LLM calls, with the model name and a preview of the prompt and response.

  • Tool calls, with the tool name, the arguments passed, and the result returned.

  • Peer delegations, showing the agent that received the delegated work and the response that came back.

  • Artifact notifications, showing the artifact name and version with a View File link.

  • Task completion or failure, with any error message and code.

Click an edge in the graph or a card in this list to highlight and scroll to the corresponding step. This cross-highlighting lines up a point in the graph with its detail.

Raw Event Stream

The step detail has a Toggle Event View switch that flips from the curated step view to the raw event stream. This view shows the underlying Agent-to-Agent (A2A) event records for the task, one card per event, unprocessed. Use it when you need to see exactly what crossed the event broker rather than the curated step summary.

Gantt Timeline

For timing analysis, select Performance in the task header to open Performance Insights. The Gantt timeline populates after the task completes, and it contains:

  • Overall Summary—the total task duration.

  • Agent Breakdown—a sortable table of each agent's total active time, LLM time, and tool time.

  • Detailed Agent View—select an agent in the table to reveal its Agent Timeline, a Gantt chart of that agent's LLM calls, tool calls, and peer delegations laid out against a seconds axis, color-coded by kind.

Tracing a Task End to End

To follow a task from the request to the result:

  1. Select the task in the session list.

  2. Read the flow graph to see the overall structure: which agents took part, where work was delegated, and which tools were called.

  3. Click through the nodes, or scroll the linear step view to inspect each step's inputs and outputs. Clicking a node highlights its step, so you can move between the visual and the detail without losing your place.

  4. Open Performance Insights to see where the time went after the task has finished.

Viewing and Downloading Artifacts

Artifacts produced during a task (files written by a tool or an agent) are reachable from the steps that created them.

  1. In the linear step view, find an artifact step. It shows Artifact: <name> (v<version>).

  2. Select View File. The Agent Mesh UI switches to the originating chat session and opens the Files tab with that artifact previewed at the referenced version.

In the Files tab:

  • When an artifact has more than one version, a Version drop-down list lets you switch between Version 1, Version 2, and so on. The list selects the latest version by default.

  • Select Download to save the current version locally.

For more information about how artifacts are versioned and stored, see Artifacts.

Sharing a Session Link

You can share a read-only snapshot of a conversation so a collaborator can review it without re-running the task. This feature is available only when your deployment has chat sharing enabled.

  1. Open the conversation you want to share and select Share.

  2. In the Manage Access dialog, either add a specific user by email as a Viewer, or select Copy Sharing Link to create a link that anyone in your organization can open.

  3. Select Save.

The recipient opens the link to a read-only view of the conversation, including its files and history, marked with a Viewer badge and the date of the snapshot. They cannot change the conversation, but they can select Continue in New Chat to fork it into their own session.

For more information about sessions and conversation history, see Sessions.

Next Steps

You have just learned how to monitor and trace task execution. Next, you might want to organize your work into reusable workspaces. For more information, see Managing Projects.