Sapient Studio

Run language models entirely on your iPhone, iPad or Mac. Chat, test API requests, benchmark speed and serve models to your other apps.

Sapient Studio runs language models directly on your device. Models download once and then work offline. Your prompts, replies and chats stay on your device: there are no accounts, no analytics and no cloud.

It is built for developers and anyone curious about running AI locally.

CHAT • Talk to small open models such as SmolLM2, Qwen, Llama and Gemma. • Replies stream live, with speed and timing under each one. • Set a system prompt per chat, and temperature and sampling per model. • Long conversations keep going when they outgrow the model's memory.

PLAYGROUND • Build an API request as a form or as raw JSON and run it on your device. • See the reply, the raw response and the timing. • Ask for JSON replies or offer tools to the model. • Copy any request as curl, Python or Swift.

LOCAL API SERVER • Serve models to your other apps and devices through an OpenAI-compatible API while Sapient Studio is open. • Pair another device by scanning a QR code. • A request log shows every call, with search, filters and replay. • Live progress in the Dynamic Island.

BENCHMARK • Measure tokens per second, time to first token and memory. • Compare two models side by side. • Run a saved set of prompts across up to three models and see what changed. • Keep a history to spot differences after an update.

MODELS • Download, load, unload and delete models, with download size and memory need shown before you start. • Keep up to four models in memory for instant switching.

GOOD TO KNOW • Models are published by third parties and download directly from Hugging Face. Each has its own licence. • Small models make mistakes. Do not rely on their answers for anything important. • JSON replies and tool calls depend on the model following instructions; small models do not always manage it. • Larger models need a device with more memory.

Sapient Studio is made by OpenHorizon Labs.