Centauri is a Chrome extension that lets you execute web tasks in natural language from a floating command shell. It is optimized for fast, visible navigation and reliable summaries. Also with extra feature to add notes to your pages. Imporve the usage of LLMs and AI agents on your daily searchs.
- Marcos Hernanz
- Javier Gil
- Martí Massó
- Marc Altabella
- Claude (Anthropic): main agent/planner/summarizer for text and task execution.
- Gemini: image analysis path when the user submits a selected image.
- ElevenLabs: text-to-speech & speech-to-text
Create a .env file in the project root:
ANTHROPIC_API_KEY=your_anthropic_key
GEMINI_API_KEY=your_gemini_key
GEMINI_MODEL=gemini-2.5-flash
ELEVENLABS_API_KEY=your_elevenlabs_key
ELEVENLABS_VOICE_ID=your_voice_id
ELEVENLABS_SPEECH_PROFILE=eleven_multilingual_v2Notes:
ANTHROPIC_API_KEYis required for the core agent experience.GEMINI_*is required for image-selected tasks.ELEVENLABS_*is required for speaker playback.- Keys are injected at build time by
scripts/build.mjs.
npm install
npm run typecheck
npm run buildBuild output is generated in dist/.
- Open
chrome://extensions. - Enable
Developer mode(top-right). - Click
Load unpacked. - Select the project
dist/folder. - Pin the extension if needed.
After code changes:
- Run
npm run buildagain. - Click
Reloadon the extension card inchrome://extensions.
- Open any page (for demo:
news.ycombinator.comormail.google.com). - Press
Ctrl+Shift+Space(macOS:Command+Shift+Space) to open Centauri. - Type a prompt and press Enter.
- Watch states progress (
Planning->Executing->Summarizing-> result).
Suggested demo prompts:
Summarize the top 5 hackernews articlesGive me a summary of my last 5 unread emails
Mode behavior:
A(Agentic): full navigation/actions.C(Chat): read-only DOM chat response (except hardcoded HN/Gmail flows that stay deterministic).
Optional helper commands:
npm run test:demo:hn
npm run test:demo:gmail
npm run test:perf
npm run test:rehearsal