1 points | by theodedra 9 hours ago
1 comments
Hi HN,
I built Nano Prompt UI, a Chrome side panel that runs entirely on-device using Chrome’s Gemini Nano API.
What it does
Side panel chat UI that sits next to whatever tab you’re on
One-click “Summarize tab” (7 bullets) using the page’s text content
Multiple chat sessions with rename / duplicate / export to Markdown
Prompt templates, image attachments (downscaled in-browser), mic input, and read-aloud replies
Why I built it I wanted something like the ChatGPT sidebar, but:
no accounts or servers
safe to use with NDA’d docs or internal tools
still usable when I’m offline / on flaky Wi-Fi
How it works
Chrome MV3 extension with side panel UI
Uses chrome.ai.languageModel / Gemini Nano for all completions (no network calls)
Simple heuristics to decide when to include page context vs just answer the question
Sessions + settings (temperature, top-K, custom system prompt) are stored locally in chrome.storage
No backend; I don’t run any servers and nothing is sent off-device
Limitations
Requires a recent Chrome build with the on-device model enabled (details in the README)
Only tested on desktop Chrome so far
Context/window size is whatever Chrome exposes for Nano right now, so huge pages get truncated
Repo (with install + flags): https://github.com/theodedra/nano-prompt-ui
I’d love feedback on:
UX of the side panel vs popup
The way I’m handling page context (when to include tab text, when not to)
Any ideas to keep it privacy-first but more powerful (better summaries, code workflows, etc.)
Happy to answer implementation questions in the comments.
Hi HN,
I built Nano Prompt UI, a Chrome side panel that runs entirely on-device using Chrome’s Gemini Nano API.
What it does
Side panel chat UI that sits next to whatever tab you’re on
One-click “Summarize tab” (7 bullets) using the page’s text content
Multiple chat sessions with rename / duplicate / export to Markdown
Prompt templates, image attachments (downscaled in-browser), mic input, and read-aloud replies
Why I built it I wanted something like the ChatGPT sidebar, but:
no accounts or servers
safe to use with NDA’d docs or internal tools
still usable when I’m offline / on flaky Wi-Fi
How it works
Chrome MV3 extension with side panel UI
Uses chrome.ai.languageModel / Gemini Nano for all completions (no network calls)
Simple heuristics to decide when to include page context vs just answer the question
Sessions + settings (temperature, top-K, custom system prompt) are stored locally in chrome.storage
No backend; I don’t run any servers and nothing is sent off-device
Limitations
Requires a recent Chrome build with the on-device model enabled (details in the README)
Only tested on desktop Chrome so far
Context/window size is whatever Chrome exposes for Nano right now, so huge pages get truncated
Repo (with install + flags): https://github.com/theodedra/nano-prompt-ui
I’d love feedback on:
UX of the side panel vs popup
The way I’m handling page context (when to include tab text, when not to)
Any ideas to keep it privacy-first but more powerful (better summaries, code workflows, etc.)
Happy to answer implementation questions in the comments.