Skip to content
FidelicRoster →

Gmail integration · OAuth + REST

Connect AI agents to your Gmail inbox

AI agents that read the Gmail labels you share, draft new messages, and post the result to Slack — inbox triage, follow-up briefs, and signal-extraction from threads.

How a Fidelic agent reaches Gmail.

The connection

What's wired

Fidelic agents connect to Gmail through OAuth + the Gmail API. Your agent reads the labels the workspace admin explicitly grants — never your full inbox by default. Scopes are read-only unless you grant draft-creation rights.

When the agent reads, it stays inside the labels you've shared. Nothing leaves your Gmail unless the agent posts a summary to Slack or creates a draft in your own outbox for human review before send.

→ Gmail official API docs

Set it up

How to connect

Before you start

  • A Google Workspace account — any plan, including individual Gmail.
  • Workspace admin or an admin who can approve third-party app access.
  • A Slack workspace where the agent will live and post.
  • A decision on which Gmail labels the agent should reach.
  • About five minutes for the OAuth flow.

The steps

  1. Sign in to Fidelic and pick the agent. From the Roster, choose the agent that needs Gmail access. Each agent has a Gmail connect button on its setup screen.
  2. Authorize Gmail. Click Connect Gmail. The OAuth dialog opens; sign in with your Google Workspace account. The app appears in Workspace Admin → Security → API controls → App access control.
  3. Pick the labels the agent should reach. Pick the Gmail labels the agent will read. Read-only by default; grant draft-creation only if you want the agent to write outbox drafts you'll approve before sending.
  4. Confirm and finish. The agent runs a test read against the first label. When it passes, your Slack channel shows a Connected to Gmail message naming the agent.
  5. Try the first agent action. Ask the agent to triage a label in Slack — it returns a summary with citation links back to specific Gmail threads.

The edges

Connection limits

Where the connection draws the line:

  • · The agent only reads labels the admin explicitly grants — no full-inbox sweeps.
  • · Drafts created by the agent stay in your outbox; the agent never auto-sends.
  • · Attachments above 25MB need to be read via download links; the agent does not host email attachments.

The path

Where the data goes

The connection runs inside your sandboxed Anthropic project. One project key per Fidelic customer; Fidelic infrastructure does not see your platform data.

Anything the agent reads stays inside your environment. The only data that leaves is what the agent posts to Slack or writes back to a destination you authorized — and you can revoke either at any time from the platform's connections panel.

→ Read the full security architecture

Where to next