Prism – AI Playground

Offline AI in your pocket. Prism is a native AI workspace for iPhone, iPad, and iPhone Duo. Run models on your own hardware with Apple Intelligence or Prism’s custom llama.cpp engine. Download models directly in Prism or import compatible GGUF files. Your conversations, model files, and settings stay on your device. No account, subscription, or cloud inference service required. Control your inference Use Metal GPU acceleration with supported models and configure how they run. Adjust context length, response limits, batch size, temperature, Top-P, Top-K, and repeat penalty. Define your own system prompt and inspect prompts in Developer Mode. Explore local models Browse a curated library of compact text, reasoning, and vision models. Download models in the background or import compatible GGUF files from Files and HTTPS URLs. Attach images with supported vision models. Measure performance Track generation speed in tokens per second, token counts, and generation duration. Review performance trends and compare runs across your local models. Keep your work close Search and reopen saved conversations, review formatted responses and inline reasoning, and export transcripts as Markdown or JSON. Thinking duration is shown for supported models. Built for Apple devices Work in a focused native interface with layouts adapted to iPhone, iPad, and iPhone Duo. Available in English, German, and Turkish. Once your chosen model is downloaded and ready, inference works offline. Internet access is required for model downloads and catalog updates. Prism does not collect analytics or telemetry. Requires iOS 26 or iPadOS 26 or later. Apple Intelligence requires a supported device and must be enabled. Model compatibility, image support, and performance depend on the selected model and available device memory.