Xprompt: Foundation Model Lab

Xprompt is a lab for Apple's Foundation Models. Write a prompt, tune it, and see exactly how the model responds, before it ever ships in your app.

TWO MODELS, ONE PLACE
Run prompts on the on-device model, or on Apple's server model through Private Cloud Compute, with a larger context window and deeper reasoning. Switch per chat, any time.

COMPARE SIDE BY SIDE
Split any chat in two and send every prompt to both. Try different models, temperatures, sampling or reasoning levels, with timing and token counts under each answer.

SEE WHAT THE MODEL SEES
• Exact input, output and reasoning token counts
• A live meter that shows how full the context window is
• The model's reasoning, streamed as it thinks
• Clear errors when a prompt is refused, rate limited or runs out of context

ITERATE FAST
• Edit any message and resend it, or branch a conversation into a new chat
• Change the instructions mid-chat; the next reply uses them
• Attach images and ask about them
• Temperature, sampling and reasoning controls
• Markdown with code blocks and tables

BUILT FOR iPAD AND MAC
A native sidebar, keyboard shortcuts, Dynamic Type and VoiceOver throughout. Available in English, Portuguese, Spanish, Italian and Chinese.

No accounts and no API keys. On-device chats never leave your device, and Private Cloud Compute is Apple's privacy-first cloud.

Requires Apple Intelligence. Image prompts, reasoning and Private Cloud Compute need iPadOS 27 or macOS 27. Private Cloud Compute has a daily usage limit set by Apple.

PART OF CAIO'S CREATIVE STUDIO
Xshirt, Xprompt and Xspeak are three apps for iPad and Mac by Caio Agra Lemos, made for the ideas you put out.