Security and data handling
Clients ask how their data is handled before they give you access to their systems. This page answers the common questions. For the full documents, use the Trust Center and the privacy policy.
Compliance
Section titled “Compliance”- superglue holds a SOC 2 Type 2 report. Request it in the Trust Center.
- superglue acts as a processor for the client data that you work with. A Data Processing Addendum covers that processing.
- Security questions go to security@superglue.ai.
Where data is processed
Section titled “Where data is processed”- superglue Cloud runs on AWS.
- On the Enterprise plan, superglue can run in a dedicated deployment, including in the EU, or on the client’s or your own infrastructure.
- superglue Cloud connects to client systems from fixed IP addresses that the client can allowlist. See Cloud networking.
- Systems inside a private network connect through the secure gateway. The gateway opens an outbound encrypted connection, so the client adds no inbound firewall rules.
AI models
Section titled “AI models”- superglue Cloud accesses AI models through AWS Bedrock. It does not send client data directly to model developers. See AI models in the privacy policy.
- superglue does not use client data to train AI models. Data sent through AWS Bedrock is used for inference only.
- Saved tools run with no AI at execution time. A scheduled sync or a migration load sends no data to a model.
Secrets
Section titled “Secrets”- Secret values are encrypted with AES-256. Stored secret values and the values that you enter in the secure form never appear in agent prompts.
- When the agent creates credentials in its browser, such as a client secret, it reads the value on the page to save it to the connection.
- The agent asks for secret values in a secure form. Do not paste them into the chat.
- Each consultant connects their own email, chat and meeting sources. The agent reads a source only with the credential of the person who asks.
Writes to client systems
Section titled “Writes to client systems”- The agent reads without asking only for reads it recognizes, such as GET requests and single SELECT queries.
- By default, the agent asks for permission before it creates, changes or deletes data, and superglue refuses writes that nobody approved. See Implementation lifecycle.
- The agent never joins or records meetings.
Audit trail and retention
Section titled “Audit trail and retention”- Every run of a tool is recorded with its trigger, user, systems and result. See Monitoring runs.
- Full step results are kept only when an admin turns on Store run results. It is off by default.
- Admins can delete all run history, stored run results and run output files of the organization with Delete all runs in Settings. This cannot be undone.
- The retention periods for each kind of data are in the privacy policy.
Access control
Section titled “Access control”- On the Enterprise plan, access rules give each role exactly what it needs, and single sign-on signs users in through the client’s or your identity provider, with SAML or OpenID Connect. See Role-based access control and Single sign-on.
- The subprocessors are listed in the privacy policy and the Trust Center.