Updating the AI model
Periodically we ship a new version of the AI model that generates your notes. When that happens, Confidant downloads the new one on its next launch.
What you'll see
On launch (with network access on), Confidant checks if a newer model is available. If yes, you get a dedicated screen:
Improved AI model
We've updated the AI model to generate better clinical notes. Downloading now — this only happens once.
A progress bar shows the download size and percentage. After download, a brief "Verifying…" step confirms the file matches what we published. Then Continue appears.
Click Continue to start using the new model. Your existing notes and patients are unaffected — the model swap doesn't touch any data, just the AI's brain.
Why model updates are bigger than app updates
App code is small (tens of MB). The AI model is the actual neural network weights — about 3 GB for the Pro tier. We don't ship model updates often (once or twice a year, typically) because they're a meaningful download for users on slow connections.
When we do ship one, it's because we've measurably improved note quality, fixed a clinical-voice issue, or addressed a real-world failure mode that the prior model couldn't handle.
Can I skip a model update?
You can quit during the download — it'll re-prompt on next launch. There's no skip button on purpose: a meaningfully-newer model is almost always better than the prior one, and we don't want users on stale models to hit issues that have already been fixed.
If you have a strong reason to stay on a specific model version (e.g. you're mid-engagement with a specific clinical workflow and want stability), contact support.
Verification
The model file is signed and verified the same way app updates are. If the SHA-256 of the downloaded file doesn't match what we published, the install is aborted and you stay on the previous model. This protects against tampered or partial downloads.
Old model files
Once the new model is verified and live, the old model file is removed to free up disk space. You don't need to clean anything up manually.
What if the model download fails repeatedly?
Usually network or disk space:
- Slow / interrupted connection. The download is resumable on next launch, but a very slow link can time out. Retry on a faster network.
- Out of disk space. The model is ~3 GB, plus temporary space for verification. Free up some space and retry.
- Antivirus quarantining the file. Some antivirus tools flag large downloads from unknown sources. Whitelist Confidant in your antivirus settings if this keeps happening.
If retries don't work, contact support with any error message you see and your OS version.