Advanced Mode

Versions & detailed logs

Two observability features visible on the page of a deployed agent (/deploy/[id]). One goes back in time, the other shows you exactly what happens during an execution.

Version history

Every time a blueprint is saved — the first save from the Builder, a redeployment, an update — we take a snapshot, persisted server-side. The Version historysection on the agent's page lists the last 20 snapshots with their date and a label.

Restore a version

Each row has a Restore button that loads the version into the Builder (dropped into sessionStorage then navigation to /builder). The existing chat is cleared to avoid an inconsistent conversation with a replaced blueprint.

Fork rather than in-place restoreRestoring does not rewrite your deployed agent. Instead it creates a draft in the Builder. If you then want to replace the agent in production, redeploy from the Builder — Forgent will give you a new blueprint id (a fork).

Visible in both modes

History is useful to everyone, so it shows up in both Simple and Advanced. The toggle doesn't hide it.

V1 limitations

  • Stored in Postgres and tied to your account — history follows you across devices and browsers
  • Max 20 versions kept per agent (the oldest ones are deleted)
  • Restoring imports name, description, model, system prompt — the tools and MCP servers stay managed by the Builder in the next conversation

Detailed logs (Advanced only)

During a task's execution, the standard activity timeline shows a friendly version of the events ("Read contract.pdf", "Generated the report"). In Advanced Mode, an additional Detailed logs section appears just below it with the raw SSE stream from Managed Agents.

What you see

  • The event counter in the header (e.g. "47 events")
  • One collapsible row per event: timestamp + event.type
  • When expanded: the full JSON of the event (tool_use_id, arguments, content, etc.)
  • A Copy all button that exports the entire log to the clipboard

What it's useful for

  • Debugging — understanding why an agent called the wrong tool or misread an instruction
  • Trust— seeing exactly what happens when a client asks "what is the AI doing with my data?"
  • Support tickets — pasting the log into a bug report is far more useful than a natural-language description
Sensitive dataThe log contains everything the agent saw and did, including the content of the files it read. Avoid publicly sharing a log from a task with confidential data — anonymize it first.