Skip to content

Security at OpenTag

OpenTag works only in the channels you invite it into and uses the access of the person asking. Gmail sends require explicit approval. Other supported operations may run immediately. Read-only connections prevent writes.
The same rules apply to every model a task goes to.

Slack Marketplace approvedBacked byY CombinatorSOC 2 Type I completed

Access and approvals

# growth

OpenTagAPP9:40

joined #growth. Added by Sofi Alejandra.

Sofi Alejandra9:41

@OpenTag what's the latest in #finance?

OpenTagAPP9:41

Working

Your data

  • Kept separate

    Each workspace runs in its own isolated environment. Memory, files and learned formats never cross between customers.

  • Never used for training

    Your messages, files and connected data are never used to train models, by OpenTag or by any provider it routes to. This is in our contracts.

  • Yours to delete

    Delete any stored context, or all of it, at any time. Deletion requests complete within 30 days.

  • Encrypted

    TLS 1.3 in transit and AES-256 at rest. Integration credentials sit in a separate vault with per-tenant keys.

When a model gets it wrong

Any model can make a mistake. OpenTag cites its sources, names the model it used and shows its plan, so you can check the work or stop it.

If a tool returns an error, OpenTag stops and tells you. It doesn't retry sensitive actions on its own.

# eng-oncall

Jordan Lee2:14

@OpenTag roll back yesterday's pricing change in GitHub

OpenTagAPP2:14

The GitHub tool returned a permissions error, so I stopped. Nothing was changed.

Used GPT-5.5 · 1 source

Documents and compliance

AICPA SOC for Service OrganizationsSOC 2 Type I completed.

Prescient Assurance examined OpenTag’s controls relevant to Security as of September 23, 2026. The report is available by request. SOC 2 Type II is in progress.

Visit the trust center