Offline PWA chat powered by Chrome's built-in Gemini Nano (Prompt API) and Microsoft Edge's Phi-4. No server, no API key, no internet required after setup.

"Visit the page once using Wi-Fi, and then use it even without Wi-Fi"

Requirements

• Chrome 148+
• Enable the flag: chrome://flags/#prompt-api-for-gemini-nano
• ~4GB free storage (for the Gemini Nano model download) (~2,4GB in the case of Microsoft Edge)
• Compatible GPU and CPU (checked automatically on first launch)
• More info about requirements here

Features

• Runs fully offline after the model is downloaded (~4 GB, one-time)
• Chat history saved locally in your browser
• File attachments, images, PDFs, text files, and code files
• Export / Import chats.
• Markdown rendering with syntax highlighting
• Customizable AI name, your name, and system prompt

How to use

1. Open the web in Chrome
2. On first launch, ChromeChat will check if Gemini Nano is available and guide you through the model download
3. Select your preferred languages (English, Spanish, Japanese)
4. Start chatting

How data is stored

ChromeChat uses two browser storage mechanisms, both completely local:

LocalStorage: stores chat history (titles, messages, timestamps) as JSON. Limited to 5 MB, so if it fills up, the oldest chats are automatically deleted to make room.

IndexedDB: stores image and file attachments as binary data. Used instead of LocalStorage because attachments can be large and would quickly hit the 5 MB limit.

When you export a chat (.crchat file), attachments are pulled from IndexedDB and embedded into the JSON file so the export is self-contained. When you import a .crchat file, attachments are extracted and saved back into IndexedDB.

Nothing is ever sent to a server, all data stays in your browser.

Troubleshooting

API not found / LanguageModel API unavailable

• Make sure you are on Chrome 148+
• Go to chrome://flags/#prompt-api-for-gemini-nano and set it to Enabled, then relaunch Chrome
• Go to chrome://components and make sure Optimization Guide On Device Model is present and up to date
• Some Chromium-based browsers (Brave, Helium, etc.) expose the LanguageModel API but don't fully implement it. If you're on one of these browsers, switch to Google Chrome or check if your browser has its own Prompt API support (many of them spoof the user agent, making browser detection unreliable and frankly not worth the effort)

Model not loading

• The API was detected but is not responding, this usually means the model component is missing or disabled
• Go to chrome://components and update/make sure is present Optimization Guide On Device Model (should be in Google Chrome)
• Open chrome://on-device-internals to see the model's actual status, file path, and any errors reported by Chrome, this is the most reliable way to diagnose download or initialization failures
• If the issue persists, try relaunching Chrome or reinstalling it

Microsoft Edge: model not loading

• Edge uses its own built-in model (Phi) instead of Gemini Nano
• Go to edge://flags/#edge-llm-prompt-api-for-phi-mini and set it to Enabled, then relaunch Edge
• ChromeChat will automatically detect Edge and adapt accordingly

Still having issues?

Open an issue on GitHub with your browser version and a description of the problem.

Why ChromeChat?

• Works anywhere no Wi-Fi, no server, no cloud dependency
• Fast responses the model runs on your hardware
• Funny Funny

---

Imagine using this so your data doesnt go to online servers, but then you remember that you are using Google Chrome

More information about the Prompt API

https://github.com/webmachinelearning/prompt-api https://www.w3.org/TR/webnn/