Is it safe to connect company email to textagent.io?

Google CASA Tier 2, passed September 29, 2026

textagent.io reads and sends Gmail, so Google treats it as using restricted scopes. Before an app gets those scopes, Google requires a Cloud Application Security Assessment (CASA). Ours is Tier 2, the OWASP ASVS-based level Google requires for the restricted gmail.modify scope and the Calendar scopes.

A Google-empanelled third-party assessor completed it on September 29, 2026 and submitted the Letter of Assessment to Google that day. Google requires it again every year, and we pay for it.

What it can access, and what it can't

When someone connects a mailbox, textagent.io asks for permission to:

• Read the inbox
• Send and save drafts, only when the user tells it to
• Mark messages read
• Move a message to the trash folder, only when the user explicitly asks
• Read, add and move calendar events

It does not access contacts or files. It cannot permanently delete email.

Nothing goes out without the user's OK

textagent.io drafts replies and texts them to the user first. It sends only when the user says send. There's no fully automatic mode, and it isn't a setting an admin or user can switch on. An AI agent with inbox access shouldn't be the one approving its own actions, so the approval lives on the user's phone, in the text thread.

Sign-in and access

People connect with Google or Microsoft sign-in (OAuth). textagent.io never sees or stores an email password. Access can be revoked at any time from the Google account permissions page or the Microsoft account apps page, and it ends immediately on the provider's side. Google Workspace and Microsoft 365 admins can also allow or block third-party apps like textagent.io with the admin controls they already have.

How data is protected

• TLS encryption in transit
• Access tokens and cached data encrypted at rest with AES-256-GCM
• Email contents are never written to operational logs
• Operational logs (timestamps and error traces, never message content) expire after 30 days
• A user can text "delete my data" and everything is erased

AI and your data

The reasoning runs on Anthropic's Claude models through AWS Bedrock. Your email, calendar and conversations with the agent are never used to train any AI model, by us or by Anthropic. Data from Google Workspace APIs is used only to provide the features the user sees.

Who else touches the data

Only the providers needed to run the service, under contract:

• Amazon Web Services for hosting and storage
• Anthropic for the AI model, via AWS Bedrock
• Twilio for text message delivery
• Stripe for payments

We don't sell data, and none of these providers may use it for their own marketing. Full details are in the privacy policy.

What we don't have yet

There's no SSO, no DPA and no admin console today. Each person signs up and connects their own mailbox. If your approval depends on one of these, email [email protected] and tell us which one. That's how we decide what to build next.