Watching a run: work log, plan and telemetry

A run is everything RoGenie does to answer one message. RoGenie shows its work as it goes, so you always know what it changed and why.

The plan checklist

For multi-step work, RoGenie writes a plan first. The checklist shows each task and ticks it off as it finishes, with a progress bar above the message box so the progress stays in view while you scroll.

The work log

Under each reply, the work log lists every action RoGenie took: what it asked Studio or a tool to do (the request) and what came back (the result). Typical steps are reading scripts, editing code, creating instances, searching the Toolbox, taking screenshots and running playtests.

Script edits show a small diff of the lines that changed.
Screenshots from Studio and playtests appear in the log; click one to zoom.
Failed steps are marked. A failed step does not always mean the run failed: if RoGenie recovered, the header says how many steps failed and that RoGenie carried on.
Provider routes. If a model route failed and RoGenie retried on another one, the log says so. Routes are shown as Provider 1, Provider 2 and so on.

You can collapse the log once a run is done. To hide tool calls entirely, turn off Enable working panel in Agent settings.

Thought summaries

Some models share a summary of their reasoning as they work. RoGenie shows these thought summaries live. They are only the summaries the model provider exposes, never hidden raw chain-of-thought. If the provider redacted a summary, the log says so. You can turn summaries off in Agent settings.

Sub-agents and background analysis

When RoGenie splits work across sub-agents, a Parallel sub-agents panel shows each one, its role, its progress and its result. Background script analysis panels show checks RoGenie runs on scripts in the background. Both are summarized above the message box while they run.

Cards that need you

Some steps wait for you:

Questions. RoGenie asks a question with options; pick one or type your own answer.
Publish approvals. Before a real upload to Roblox, RoGenie asks for approval. See Publishing to Roblox.
Plan approvals. In Plan mode, the plan waits for Approve & build.

How a run ends

The header of each reply shows how the run ended:

Label	Meaning
Working...	The run is still going.
Worked	The run finished.
Stopped	You pressed Stop. Everything that already ran in Studio is kept; nothing was undone.
Interrupted	The run was cut off before RoGenie saved a final reply, for example by a connection drop or a server restart. The completed work and commentary are kept in the log.
Partly done	The run ended with an error after doing real work. The completed steps are in your place.
Errored	The run failed before doing useful work. The card says why and offers Regenerate and Switch model.

If a run stopped because your quota ran out, the reply says so and RoGenie can pick up from the place's current state once your quota frees up. See Quota messages and model errors.

Info: the telemetry of a reply

Click Info on a reply to see its telemetry:

Field	What it shows
provider	Which route answered, shown as Provider N.
model	The model that ran.
mode, effective_mode	The pacing mode asked for and the one used.
prompt_tokens, completion_tokens, reasoning_tokens, total_tokens	Token counts. For a run with several model calls, the total covers the whole run.
quota_billable_tokens	What your quota counted for the run, after the model's multiplier.
model_calls	How many model calls the run made.
context_used_tokens	How much chat memory the last call used.
thinking_budget, max_output_tokens, temperature	Generation settings.

Notifications and sounds

RoGenie can play a short sound when a reply finishes and show toasts when Studio or the connection connects or disconnects. Both are switches in Settings › Appearance › Effects and sounds.