Pocket Pi support.

Your first model

1. Open Pocket Pi and choose what you want to do in the welcome flow: chat, understand photos, create images, or keep the download small.
2. Review the suggested model, download size, required components and free storage. Confirm when you are ready to download.
3. Keep a reliable Wi-Fi connection for the first download. Public suggested models do not need a Hugging Face account or token.
4. When a chat model is installed, open Chat, enter a short question and tap Send. A downloaded image model is managed in Create.

Pocket Pi requires iOS 26 or later. Hardware-aware suggestions help you start; they are not a promise of answer quality or speed on every device. AI features are free, including model selection and image creation.

A download is slow, paused or unavailable

Open Models to review chat and photo-model transfers. Open Create to review the image model. Check free storage, your network connection, and whether Offline only is enabled in download settings.

Downloads use Wi-Fi unless you enable mobile data. Large files can take time. Downloads can continue when you leave the app; if you force-quit, reopen Pocket Pi to resume. Review the download's status and available retry or discard action rather than starting a second import.

Each required file is checked before use. If a file cannot be verified, the model will not be treated as ready. Keep enough space for download staging as well as the final model; the review screen shows the app's storage estimate.

A Hugging Face read token is optional for private or gated models you deliberately choose. Any model license or publisher access requirement still applies. Never email a token to support.

Model search sends the search text to Hugging Face, and model browsing or downloads send repository/file requests and ordinary network information. A configured token associates authenticated requests with your Hugging Face account. These requests do not include your AI chats or source contents. Read the model-provider privacy explanation before adding a token.

Chat and photo understanding

If Chat says a model is needed, return to Models and select an installed chat model. Wait for active work or a model transition to finish, then try a short prompt. Keep your draft while checking readiness.

For questions about a photo, choose a photo-capable model and install its matching photo component. Text-only chat weights cannot see an image. OCR can supply extracted text, but it does not understand the whole visual scene.

Select sources explicitly for the current chat. A project can keep reusable materials together; that does not silently attach every project file to every prompt. Review the attached sources and the answer's coverage.

Model outputs can be incomplete or incorrect. Smaller prompts, fewer attached sources and a lighter model can help when memory, temperature or context limits interrupt a request.

Create an image

Open Create, review the separate image-model download, and allow approximately 4 GB for its installed assets plus additional storage for staging and saved images. A chat model can help develop a description, but is not required to generate an image.

You can develop an image idea in Chat and open it in Create for review. Edit the description there before generating; opening the idea does not automatically send a prompt or start a render.

Keep Pocket Pi in the foreground during generation. Leaving the app stops the run and preserves your description. If the app reports a memory or temperature limit, let the device cool, close the current task and try again with the available settings. Available models and output sizes depend on device capability.

Save a chosen result to Photos, or use Preview & Markup to create a separate edited copy. The original remains available in Pocket Pi until you delete it. Outputs and model performance vary; image generation is intended for a supported physical device.

Ads and the one-time Remove ads purchase

All AI features are free, even if you decline advertising or tracking. Optional non-personalized banners require a checked ad-removal status, your choice to continue with ads, authorized Apple tracking permission, and applicable Google consent. They can appear only while browsing eligible main-app screens. New ad requests pause during Offline only mode or active app work. Non-personalized ads can still use identifiers for advertising measurement across apps; see the privacy policy.

Open Settings → Ads & purchases → Remove ads. The purchase screen shows Apple's current localized price. The intended US price is US$0.99; there is no subscription.

Can I decline tracking and still use AI?

Yes. Choose Ask App Not to Track when Apple asks. Denied or restricted tracking permission keeps the Google Mobile Ads SDK from starting and blocks new ad requests. Every AI feature stays free; no purchase is needed to decline. Apple controls permission in Settings → Privacy & Security → Tracking, where available. If you revoke permission after ads have already started, Pocket Pi removes banners and stops new requests; an in-flight SDK request may still complete.

I already bought Remove ads

Use Restore purchases on that screen while signed into the same App Store account. Apple may need a network connection to check the purchase. A successful verified restore removes advertising; no Pocket Pi account is needed.

The price or purchase status is unavailable

Use the screen's retry action, check the App Store account and connection, and try again. An unknown purchase status pauses advertising. A pending purchase is not a completed purchase; wait for approval rather than buying again. All AI features remain available.

Google advertising privacy choices are available in Settings → Ads & purchases when a required privacy-options form is provided for your region. These choices are separate from Apple tracking permission and the Remove ads purchase. Apple's purchase and refund rules are managed by the App Store.

Offline mode, native tools and your data

Offline only pauses model browsing and downloads, disables paired Mac connections, and pauses new ad requests. An advertising SDK request already in progress may still complete. Installed local models can still work. This switch does not disable the iPhone's network or independent Apple services, purchases, file providers or account synchronization.

Native tools ask for permission when needed. Reminders are created only after you review and confirm the chosen checklist. Calendar and Reminders may sync through your system accounts; exports and Photos copies are separate from Pocket Pi's stored originals.

In Apple integrations, choose whether project names are offered outside the app and which titles appear in Spotlight. These choices are off until you opt in. Existing system widget or Shortcut configurations can retain a saved name; edit or remove those configurations if needed.

Use the app's deletion and storage controls for local work. Deleted chats stay in Recently Deleted for 30 days unless you remove them permanently. Deleting a project folder does not itself erase its chats or Library sources. Delete exported copies, Apple-native items and backups separately if you no longer need them.

Read the full privacy policy for model downloads, optional Mac connections, advertising SDK data, storage and support email.

Contact the developer

A useful report includes your iPhone or iPad model, iOS version, Pocket Pi version and build, the model you selected, what you were trying to do, and the exact error. A screenshot can help if you remove anything private first.

Do not include passwords, Hugging Face tokens, pairing credentials, private documents or full chat histories. If you choose to export diagnostics, review the report before emailing it. The app's diagnostic export omits chat, document and audio content.

Support email is handled through the email provider, separately from local inference. Only send information you want the developer to receive.