ClipWhat are the three engines?
Local uses the configured local reasoning path, including a supported installed model or a local endpoint. Your API uses your configured cloud provider and is subject to account entitlement and provider billing. Rush is ClipLogger's hosted batch service. These are routing choices for reasoning; OCR, transcription, visual matching, and other tasks also have their own settings.
Which engine runs by default?
The result depends on the task, available engines, settings, and any remembered project choice. Read the current engine chooser and Settings → AI rather than assuming every run uses the same model. When several eligible families are ready, the app can ask you to choose Local, Your API, or Rush. A local-family choice pins that run to the local route.
I have more than one engine set up. Which one is used?
For eligible runs, the engine-family choice sits above the individual local or provider configuration. A remembered choice is used only while it remains available. If the intended engine is unavailable, inspect the reported result rather than assuming the app quietly switched to a different provider. Settings → AI shows the task routing and local reasoning chain.
Does the AI write metadata by itself?
Analysis can persist evidence, transcripts, proposals, and attempt history. The review workflow separates those records from confirmed logging decisions, and some kinds of proposed ranges have separate acceptance settings. Review the values and active settings before applying a batch. Avoid treating every persisted AI record as a human-approved log.
What does the AI actually look at?
ClipLogger selects frames that help a model understand the clip and includes your logging fields and enabled context, such as a transcript, text seen in the image or relevant roster details. A 20 GB camera clip does not become a 20 GB video upload. Rush can also receive audio for transcription. Review the proposed log against the original because selected frames can miss a brief moment.
What is Prepare?
Prepare Suggestions is the batch verb: it runs the configured engine over the selected or browsed set and stages suggestions for review. You can stop a run (completed clips keep their results) and retry just the failed clips. Opening a clip also auto-runs light on-device passes (OCR and transcription) by default, so search and fields have something to work with before you ever hit Prepare. Turn that off in Settings ▸ AI if you don't want it.
What is Re-reason?
Run another reasoning attempt for the current clip after improving context, changing the engine, or investigating an incomplete result. Inspect which evidence is available and review the new proposals. Confirmed decisions and attempt history have their own state; do not confuse a new proposal with an instruction to replace the reviewed log.
Does ClipLogger learn from my corrections?
Corrections improve the reviewed records and context available to your workflow. That is different from retraining a foundation model. Do not assume that confirming one athlete guarantees every later appearance will be identified correctly. Check the next proposals, especially when uniforms, lighting, teams, or camera angles change.
Do I need a GPU or a beefy Mac for local AI?
Model fit depends on unified memory, quantization, context, and the other applications running. Select a supported model that fits with enough headroom for playback and editing. The hardware page's larger-model timing examples are modeled estimates, not measured guarantees for your Mac. Evaluate a representative sample and compare both elapsed time and review effort with Rush.