eBPF Continuous Profiling released

Last updated:

Approve Write ActionsRelease historyThis feature was introduced by Approve Agent0's write actions before they run on August 14, 2026.

How to make Agent0 ask for your approval in the conversation before it runs an action that creates, changes, or deletes something.

Agent0 can do more than read your telemetry — it can send a Slack message, create an automation, or call a write tool on one of your MCP servers. Write action approval puts you in front of those actions: Agent0 stops before it runs one, shows you exactly what it wants to do, and waits for you to approve or deny it.

This is a Chat feature. It covers the Agent0 page and the Integrated Experience side panel — the two places where Agent0 can show you a card and resume the conversation after you decide. To control what Agent0 may do in an automation, use its Allowed Tools guardrail instead.

Turn It On

Write action approval is an organization-wide setting, off by default. An organization admin enables it in Settings → Agent0 with the Write action approval switch.

Write action approval toggle in the Agent0 organization settings

The switch applies to everyone in the organization, and turning it on or off is recorded in your audit log.

What Needs Approval

Covered write actions include sending a Slack message, creating or updating an automation, and write tools on connected MCP servers.

Reading your telemetry never asks for approval, so investigations, queries, and analysis are unaffected. GitHub and git actions are not covered by this setting; they are governed by the permissions of the GitHub connector itself.

Creating an artifact proposal, maintaining your private memories, and attaching media to your thread do not require a write-action approval. Creating a proposal does not apply it to the underlying resource.

Approve or Deny

When Agent0 reaches an action that needs approval, it stops there and shows a card in the conversation. The card names the action, gives Agent0's reason for wanting to run it, and lets you expand the arguments to review their exact values before anything happens.

  • Approve runs precisely the call shown on the card, then Agent0 continues from its result.
  • Approve for this thread also approves other pending calls of the same tool and your later calls of it in this thread. Later calls can use different arguments. This allowance belongs to you: another person continuing the thread must approve the tool for themselves.
  • Deny tells Agent0 the call was rejected. It carries on with the rest of your request without it.

If Agent0 requests several approvals, decide each pending card before the conversation continues. Approving a tool for the thread also resolves its other pending cards. Tools you have already approved for the thread run without another card.

There is no control to withdraw a thread-scoped approval. Start a new thread if you want Agent0 to ask again for that tool.

The Conversation Waits for You

An approval is not a prompt you have to answer immediately. The conversation pauses where it is, so you can close the tab, check with a colleague, and come back to approve or deny later.

Choosing Approve authorizes one call. Reloading the page or repeating the click cannot run that call again. Approve for this thread permits later calls of that tool as described above.

What Gets Recorded

Chat Details logs the full lifecycle of every approval as part of the conversation's activity:

  • Requested approval: {action} — where Agent0 stopped
  • Action approved: {action} or Action rejected: {action} — what you decided
  • Action approved for this thread: {action} records a thread-scoped approval

Chat Details reasoning steps showing a requested approval, the approval, and the tool call that followed it

Dash0 also stores who approved or denied each action and when, so an approved write action always traces back to a person.

Further Reading