System requirements

• iOS 17.0 or later
• iPhone or iPad with Apple Silicon (A14 or newer; M-series iPads recommended)
• ~3 GB free storage per model you want to keep
• At least 6 GB of device RAM recommended for 7B-class models

Troubleshooting

"No model loaded" on the Chat tab

You need to download and play a model before chatting. Go to the Models tab (Record Cabinet), tap Download on any model, wait for the download bar to complete, then tap Play. Switch back to Chat and it should be ready.

A model finished downloading but won't load

Larger models (7B class, ~4 GB) need at least 6 GB of device RAM. If load fails on an older device, try a smaller model like Llama 3.2 3B Instruct (~2 GB) or Phi-3.5 Mini (~2 GB).

Download stopped mid-way

Tap the model card again — downloads resume from where they left off. If a URL has gone stale upstream, please file a GitHub issue with the model name and we will refresh the catalog.

Restoring purchases on a new device

Settings → Restore Purchases. The button is in the paywall when your eval window has ended, or in the StoreKit recovery flow on first launch on a new device. Same Apple ID required.

Refunds

App Store purchases are refunded by Apple, not by us. Request at reportaproblem.apple.com.

How do I use the optional local server?

Settings → toggle the server on, generate an API token, then point your tooling at http://<device-ip>:11434/v1/chat/completions with the bearer token. The server is disabled by default and only listens on the local network (or loopback if you set Listen Address to 127.0.0.1).

My eval period expired — can I get more time?

Email us and we will sort it out.

---

About this app

Bard LLM runs open-weight large-language models entirely on your device using Apple Silicon's Neural Engine and GPU via llama.cpp. Nothing you type leaves your phone. There is no account, no analytics, no third-party SDKs. Full details on our privacy policy page.