A developer released seed., a self-modifying webpage that uses Chrome's built-in `window.LanguageModel` API to run Google's Gemini Nano entirely on-device, requiring no API key or network calls. The tool, which stores the entire site within the URL, requires Chrome 127+ with a specific flag enabled and downloads a 1–2 GB model on first use. The project highlights the potential and limitations of on-device AI, as Gemini Nano may struggle with complex prompts but performs well on short, focused requests.
Uses Chrome's built-in window.LanguageModel
API (Gemini Nano, runs entirely on-device). No API key, no network calls.
checking availability…
Setup: Chrome 127+ with the flag chrome://flags/#prompt-api-for-gemini-nano
enabled. The model downloads (~1–2 GB) on first use.
Note: Gemini Nano is small and may struggle to produce well-formed HTML for complex prompts. Best results on short, focused requests.
EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.