Codeuctivity inference service

PromptForge Beta

API checking Agents checking

Settings

Client API key

The key stays in this browser's local storage and is sent only to this origin.

Model

All browser options download and run the model on this device the first time they're used; no prompt, image, or completion is sent anywhere. Gemma 3 270M, Qwen3-0.6B and Florence-2 need no API key. The larger models (Qwen3-1.7B, Gemma 3 1B and the DeepSeek models) additionally require a valid client API key to unlock (checked once, never sent with a prompt). Florence-2 reads an uploaded image instead of chatting - select it to attach a photo for OCR. Other, remote models require a client API key and a connected inference agent.

Temperature

Maximum output tokens

A safety limit, not a target length: answers longer than this are cut off.

System prompt (optional)

Ask something to get started.

The browser model above runs entirely on this device - no key, no agent.

Firefox on Android and Safari on iOS/iPadOS have known reliability issues running this model in-browser; Chrome or Edge is recommended if you hit failures.

The in-browser model isn't available on iPhone - a WebKit memory limit makes it crash and reload the page every time, on every iPhone browser. Add a client API key in Settings to use a connected model instead.

Health Readiness Version

Download the local model?

Running the browser model downloads the Gemma 3 270M model (roughly 570 MB) to this device. It downloads once and is cached here afterward; no prompt or completion is ever sent anywhere.

Firefox on Android and Safari on iOS/iPadOS have known reliability issues running this model in-browser. Consider Chrome or Edge instead if it fails.

Welcome to PromptForge

Two different ways to try a prompt, right on this page:

Free, in-browser Gemma: Gemma 3 270M runs entirely inside this browser tab - no API key, no server round trip, nothing you type ever leaves your device. It's free for anyone to try, and it's the part of PromptForge currently in public beta.
Bring your own key: With a provisioned client API key, PromptForge can instead dispatch your prompts to a connected inference agent running a full-size model. This path isn't part of the public beta - it needs a key that's issued directly, not handed out publicly.

This won't show again - it's remembered in this browser's local storage.