Getting Started

Start with Cordy for Chrome 2.4.x: chat, browser context, tab and bookmark libraries, translation, and local or cloud models.

Cordy for Chrome is an AI assistant that lives inside your browser. It opens in the Chrome side panel, reads the page you're on (only when you ask), acts on text you select, and can search and organize your tabs, history, and bookmarks — all driven by the AI model you choose.

Cordy is bring-your-own-key (BYOK) and browser-direct: you supply an API key for a provider you already use, and Cordy talks to that provider straight from your browser. There is no Cordy server in the middle, no account, and no telemetry. Your keys are encrypted on your device, and your conversations stay on your device.

Cordy for Chrome was formerly called Web Nexus. Source and tag are at 2.4.5; the latest published GitHub Release is 2.4.4. The repository is private, so downloading that release requires repository access.

What you can do with it

  • Chat in the side panel about anything, with streaming replies, multiple parallel conversation tabs, message branching, and attachments. See Chat.
  • Act on selected text on any page — translate, explain, summarize, quote into chat, or read aloud — from a floating menu. See Floating menu.
  • Run commands fast with an in-page command palette: web search, jump to Chrome pages, switch tabs, open bookmarks, and slash commands. See Command palette.
  • Ground the AI in your browser with @-mentions of the current page, other tabs, history, and bookmarks. See Mentions.
  • Manage browser material — inspect current tabs and history, collect or batch-manage tabs, keep read-later links, and organize bookmarks with manual tags, notes, duplicate checks, and link-health checks. See Tabs, history & collections and Bookmarks.
  • Keep conversations — save, search, organize, import, and export chat history from the dashboard. See Conversation library.
  • Translate long text side-by-side in a dedicated workbench. See Translation.
  • Listen to replies and page selections with local or cloud text-to-speech. See Read aloud (TTS).
  • Run models on-device with Chrome's built-in AI, in-browser models via WebGPU, or your own local runtime (Ollama, LM Studio, and more). See Local & on-device AI.

Quick start

Cordy is not on the Chrome Web Store yet, so setup takes a few minutes the first time. The steps below are the short version — the Installation page walks through each one in detail.

1. Install the extension

Load Cordy as an unpacked/developer build or from a release archive, then pin it to your toolbar. You need Chrome 148 or newer. Full instructions are on the Installation page.

2. Open the side panel

Click the Cordy icon in the toolbar, or press the keyboard shortcut:

PlatformShortcut
Windows / LinuxCtrl + Shift + Y
macOSCmd + Shift + Y

The side panel opens on the right, docked to the current window.

3. Add a provider API key

Open the dashboard and go to Settings → AI Config → Providers, choose a provider, and paste in your own API key.

Cordy ships with seven built-in cloud providers — OpenRouter, OpenAI, Anthropic, Google AI, xAI, Vercel AI Gateway, and Volcengine Ark — plus custom OpenAI-compatible endpoints. See Providers & models for where to get a key.

Your key is encrypted and stored only on this device. Cordy never uploads it anywhere except the provider you configured it for.

4. Pick a model

Under AI Config → Feature Models, choose a model for Chat. Translation, explanation, summarization, and TTS have their own slots where applicable.

5. Send your first message

Type a question in the composer and press Ctrl + Enter (or Enter, depending on your send-key setting). The reply streams in.

The first time you send anything to a cloud model, Cordy shows a one-time consent dialog explaining that your message goes to the third-party provider you configured. Accept it once and you won't be asked again (you can withdraw consent later in Settings → Data & privacy). Local, on-device models never trigger this dialog.

Prefer to stay fully on-device?

You don't need a cloud key. Cordy can run prompts on your machine with Chrome's built-in Gemini Nano, in-browser Gemma models over WebGPU, or a local runtime such as Ollama or LM Studio. Model-weight downloads and loopback permission prompts are described in Local & on-device AI.

How Cordy handles your data

  • Your keys, your data. API keys are encrypted on-device; conversations and saved content live in local browser storage.
  • Browser-direct. Prompts go only to the provider you configured. Cordy runs no backend of its own.
  • No telemetry. Cordy collects no usage analytics and reports to no server.
  • Permissions by purpose. Broad website access, history, bookmarks, browsing-data cleanup, and loopback runtime origins are optional grants. Tab metadata and the side panel use manifest permissions installed with the extension.

The full model is described in Privacy & permissions.

Next steps