Vaiz
Sign up
Help center
Tutorials

How to review and manage agent runs in Vaiz

Every time an AI agent executes in Vaiz—whether on a schedule, in response to a workspace event, when called in a task, or started manually—it records a run. From Runs, you can inspect the agent's report, review and approve proposed changes in Ask mode, trace every tool call and reasoning step, and monitor AI credit usage.

The Runs list in Vaiz showing agent runs, statuses, action counts, and triggers

Step 1. Find agent runs

You can open agent runs from four places in Vaiz:

  • Runs in the left sidebar — shows runs from all agents across the space that you have access to.
  • Project → Runs — shows runs from agents scoped to a specific project.
  • Agents → [Agent] → Runs tab — shows the run history for a single agent (while dry-run tests stay in the agent's Test tab).
  • Inside a task — the Agent runs section in the task sidebar lists every run that created, edited, or commented on that task.

Each row in the Runs list displays:

  • Run number and agent name — for example, #142 Release Milestone Planner.
  • Status badge — Queued, Running, Awaiting review, Suggestions ready, Applied, Rejected, Failed, or Cancelled.
  • Action count — how many write actions (such as creating milestones, editing tasks, or posting comments) the agent recorded during the run.
  • Timestamp and trigger — when the run started and what caused it (for example, Triggered by schedule, Triggered by task update, Called by a member, or Run manually by You).

Step 2. Filter the Runs list

Use the filter bar at the top of Runs to find specific executions:

  • Search — filter runs by agent name.
  • Date — narrow runs to a specific date range.
  • Agents — select one or more agents.
  • Status — filter by run status (for example, select Awaiting review to see all runs waiting for your approval).
  • Mode — filter by the mode the run executed in (Auto, Ask, or Suggest).
  • Trigger — filter by how the run was started (On schedule, When a task is created, When a task is updated, When a comment is added, When a milestone is created, When a milestone is updated, When assigned or mentioned, or Manually).

Step 3. Inspect a run's Attribution and Summary

Click any row to open the run details panel on the right (or Cmd/Ctrl + click to open it as a full page).

Run details showing Attribution and Summary for a Release Milestone Planner run

At the top of the run, two sections give you the context and outcome at a glance:

  • Attribution — shows which agent ran, how and when it started, how long it took, which AI model it used, and how many Credits and Tokens (in and out) it consumed. When applicable, it also shows the skills and MCP servers active during the run, who triggered the event (Caused by), whose permissions bounded the run (Limited to the access of), and who reviewed or applied the changes.
  • Summary — the agent's structured report in Markdown, with clickable links to every task, milestone, or document it referenced.

Step 4. Approve or reject proposed actions (Ask mode)

How actions behave depends on the agent's Mode at the time of the run:

  • Suggest mode (Suggestions ready) — read-only. The agent records what it would do as suggestions without changing any workspace data. If you like the result, click Switch agent to Ask right from the run so future runs can request approval.
  • Ask mode (Awaiting review) — low-risk comments are posted immediately (marked Applied automatically), while task edits, new tasks, milestone changes, document edits, and Space Memory entries wait for your decision.
  • Auto mode (Applied) — allowed actions are applied during the run and logged on the run page for auditing.
Reviewing and approving proposed actions in Ask mode

When a run is Awaiting review in Ask mode, you can:

  1. Review actions one by one — click Approve or Reject on individual action cards (such as Create Milestone, Edit Milestone, Edit Task, or Set Blockers), then click Apply approved to execute only the approved changes.
  2. Apply everything at once — click Approve and apply all to approve all pending actions and execute them immediately.
  3. Discard the proposal — click Reject all to reject every pending action without modifying your workspace.
Note: Actions applied in Auto mode are checked by the Vaiz safety guardrail (Jev). If a tool call contradicts the agent's instructions or scope, it is held back and marked Blocked by safety check on the action card.

Step 5. Trace tool calls in Steps

Expand the Steps section at the bottom of the run to see the chronological execution trace with timestamps.

Expanded Steps section showing tool inputs, outputs, and reasoning steps

Each step shows either the agent's intermediate reasoning or a tool call (such as Get Tasks, Get Milestones, Create Milestone, or an external MCP tool). Click any tool step to expand its JSON payload and inspect the exact parameters and results.

Step 6. Re-run or cancel a run

Use the three-dot menu in the top-right corner of the run header to:

  • Run again — start a new run for manual or scheduled agents directly from the run view.
  • Open agent — jump to the agent's configuration, instructions, and Test tab.
  • Cancel run — stop a run that is currently Queued or Running.

To configure agent instructions, triggers, schedules, and allowed actions, see How to create custom AI agents in Vaiz to automate routine work.

Setting up your first AI agent?
Learn how to create custom AI agents from a plain-language description or manual setup, choose triggers and allowed actions, and explore ready-to-use prompt examples.
How to create custom AI agents in Vaiz