Skip to main content

Automate Gmail and Google Calendar with an AI Agent in Your Browser

Most "AI for Gmail and Calendar" tools ask you to connect your Google account to their servers, build a workflow in a dashboard, and trust a third party with an API token to your inbox. Vibe takes a different route: it's a Chrome extension with a side panel agent that works inside the browser session you're already logged into. You type a request in plain English — "what's on my calendar this week?", "draft a reply to the meeting invite that conflicts with my flight" — and the agent calls Gmail and Calendar tools directly from your browser, in your session.

This guide shows what that looks like in practice, using only capabilities that actually ship in the product today: Gmail search, reading and drafting, sending, Calendar viewing, event creation and deletion, and cross-app tasks that touch both — plus Google Drive, Docs, and Sheets from the same connection.

What you need​

  • The Vibe Chrome extension installed.
  • Your Google account connected once under Settings → Google Workspace in the extension. This is a standard Google OAuth consent screen — you see and approve the scopes (Gmail read/send, Calendar read/events, Drive read) before anything is granted.
  • That's the whole setup. No API project to register, no keys to paste, no workflow builder.

Once connected, the agent can call gmail_search, gmail_get_message, gmail_get_thread, gmail_create_draft, gmail_send_message, calendar_view, calendar_create, calendar_delete, and the Drive/Docs/Sheets tools from any tab. In Ask Mode the agent is read-only; in Agent Mode (the default) it can act. See Using the Co-Pilot.

1. Morning triage: inbox and calendar in one prompt​

Instead of opening Gmail, scanning unread mail, then switching to Calendar, ask for both at once:

"Show me my unread emails with a one-line summary of each,
then show my calendar events for today and tomorrow."

The agent runs gmail_search (with a query like is:unread), pulls the messages it needs with gmail_get_message, then calls calendar_view for the date range — and gives you one combined answer. Because this is read-only, it's a good fit for Ask Mode.

2. Answer "what does my week look like?" without opening Calendar​

"What meetings do I have scheduled this week on my Google Calendar?"

calendar_view returns your events for the range you describe, and the agent summarizes them — including telling you honestly when a day is empty. You can narrow it naturally: "just Thursday and Friday", "only meetings with more than two attendees", "anything before 10am".

3. Create and clean up events in plain English​

"Create a 30-minute event today at 11:00 AM called 'Design review'."

The agent calls calendar_create with the title, time, and duration you gave it. Cleanup works the same way:

"List my events for today, then delete the one called 'Design review'."

That's a two-tool chain — calendar_view to find the event, calendar_delete to remove it. One honest note: there is no dedicated edit/reschedule tool today, so "move my 2pm to 3pm" means the agent removes the old event and creates a new one rather than editing it in place.

4. Cross-app: use Calendar context to write email​

This is where a single agent beats separate Gmail and Calendar add-ons — one request can read from Calendar and write to Gmail:

"Check my calendar for tomorrow. If the 2pm meeting conflicts with anything,
draft a reply to the organizer proposing 30 minutes later."

The agent calls calendar_view, finds the conflict, reads the invite thread with gmail_get_thread, and writes a grounded draft with gmail_create_draft. The draft lands in your normal Gmail drafts folder — review, edit, or send it from Gmail like any other.

The same connection covers Drive, Docs, and Sheets: ask the agent to research a topic on the web and write the summary into a new Google Doc (docs_create_document + docs_append_text), collect data into a Sheet (sheets_create_spreadsheet + sheets_update_values), or list your recently modified Drive files (drive_list_files).

For a deeper Gmail-only walkthrough — search operators, draft-vs-send patterns, thread summarization — see How to Automate Gmail with AI.

How this differs from Zapier-style automation​

Tools like Zapier, Make, or n8n connect Gmail and Calendar through server-side API integrations: you authorize their platform, define a trigger ("new email matching X"), and a pipeline runs unattended on their infrastructure. That model is genuinely better for recurring, scheduled, always-on automation — if you need "every invoice email auto-filed to a spreadsheet at 2am", use a workflow tool.

Vibe's model is different in three concrete ways:

  • It works on your live, logged-in session. The agent acts from your browser, in your existing Google session, rather than through a third-party server holding an offline token to your account.
  • There's no workflow to build. You describe the task when you have it, in plain English. No triggers, no field mapping, no multi-step zap editor — which also means nothing to maintain when your needs change.
  • It's interactive, not unattended. You're present for every run, can watch the agent work in the side panel, and can stop it. That makes it the wrong tool for scheduled background jobs and the right tool for the ad-hoc, judgment-heavy tasks that never fit neatly into a trigger-action pipeline.

Honest limitations​

  • Sending is real. gmail_send_message sends immediately, exactly like clicking Send. If you want a review step, ask for a draft explicitly and send it yourself — the Gmail guide covers the draft-vs-send patterns in detail.
  • No unattended scheduling. Vibe runs when you ask it to, in an open browser. It does not run cron-style background jobs.
  • Non-deterministic. The same prompt can take slightly different paths or phrasing across runs. Verify anything you're relying on for a decision.
  • No calendar edit tool. Changing an event means deleting and re-creating it, as noted above.

Privacy​

The email and calendar content the agent reads is sent to whichever AI model you've selected — the managed Vibe API, your own API key (OpenAI, Anthropic, Gemini, OpenRouter, Azure), or a fully local model (Ollama or Chrome's built-in Gemini Nano). With Ollama or Gemini Nano, inference runs on-device and that content doesn't leave your machine. Your Google OAuth tokens are stored locally in your browser. See Providers for the comparison and the Privacy Policy for full details.

Get started​

Install Vibe for Chrome, open Settings → Google Workspace to connect your account, and try the morning-triage prompt from section 1. It's a two-minute setup for an assistant that reads your inbox and your calendar in the same breath.

FAQ​

Do I need to set up API access or a Google Cloud project? No. Connecting your account is a one-time Google OAuth consent inside the extension. There are no API keys to create and no third-party dashboard holding access to your account.

Can it schedule meetings for me automatically every day? No — Vibe is an interactive agent, not a background scheduler. It creates events when you ask it to (calendar_create), but it doesn't run unattended recurring jobs. For trigger-based automation, a workflow tool like Zapier is the better fit.

Can it read an email and update my calendar in one request? Yes. A single prompt can chain Gmail and Calendar tools — for example, read a confirmation email with gmail_get_message and create the matching event with calendar_create, or check calendar_view before drafting a reply to an invite.

Which AI model sees my email and calendar data? Whichever one you pick in Settings. Cloud providers see the content the agent reads; local options (Ollama, Gemini Nano) keep it on-device. See Providers.