.DkNUKaKx_ZLAT4m.webp?dpl=dpl_mxdY43eVEb2xCRXDR6YEJEsw9FBU){kind=small}
An integration platform holds credentials to every system it connects and moves data between them, so a security review should focus there. How are credentials stored and used? Who can build and run workflows? What gets logged, and where does it go? Where is data processed, and what happens to data sent to AI models? Certifications are the floor. Tray.ai is audited for SOC 1 and SOC 2 Type 2, encrypts data in transit and at rest, enforces SSO and phishing-resistant two-factor authentication, logs every action with streaming to your own systems, and hosts in the US, EU, or Asia Pacific. The rest of the review is about your configuration, because a well-secured platform can still run a badly secured workflow.
Most SaaS reviews ask what a tool stores and who can see it. An integration platform needs more, because of what it’s connected to. It holds the keys to your CRM, ERP, HR system, and data warehouse at once, and every workflow it runs reads from one and writes to another. A gap in the platform is a gap in all of them.
Why integration platforms get a different review
Credentials to many systems sit in one place. API keys and OAuth tokens for every connected system are stored by the platform and used on your behalf.
Data moves through workflows. Records pass through the platform on the way from one system to another, sometimes transformed on the way.
Logs hold business data. Execution logs show each step’s input and output, so they contain whatever the workflow moves: customer records, invoices, employee data.
Webhooks are public endpoints. A workflow triggered by a webhook can be started by anyone who can reach its URL, unless it’s secured.
AI steps send data to models. When a workflow or agent calls a language model, data leaves the workflow for the model provider.
Questions to ask any integration platform
Audit reports and testing
- Which independent audits do you complete, who performs them, and what period does the latest report cover?
- When was your last independent penetration test, and do you run a vulnerability disclosure or bug bounty program?
Credentials
- How are credentials for connected systems stored and encrypted, and are they kept out of logs?
- Can a workflow or agent act with the end user’s own permissions rather than a shared account?
Access and identity
- Do you enforce SSO and two-factor authentication?
- How are roles and permissions scoped, and can access be separated by team or environment?
- When can your support staff see our data, and is that access logged?
Logging and data in logs
- What’s logged, for how long, and can we shorten retention or stop storing workflow data?
- Can sensitive values be masked in logs?
- Can logs stream to our SIEM?
Data residency and resilience
- Which hosting regions can we choose?
- How often is data backed up, and how often is disaster recovery tested?
AI data handling
- Do you train models on our data, and do AI providers retain prompts or responses?
- Can we opt out of AI features, and control which models are used?
Inbound requests
- How are webhooks and other public triggers secured?
Contract
- Is there a standard DPA, and a current sub-processor list?
- How quickly are customers notified of a confirmed incident?
What stays yours to configure
A platform’s controls only work if they’re used. Five things sit with your team whatever platform you choose:
- Least-privilege credentials. Connect each system with an account that can do only what the workflows need.
- Authenticated webhooks. Require a token on every public trigger rather than relying on an unguessable URL.
- Log retention and masking. Keep logs only as long as you need them, and mask fields that hold personal data.
- Who can deploy to production. Separate building from publishing, using roles and environments.
- What reaches AI steps. Send models only the data the step needs.
Is Tray.ai secure enough for an enterprise to use?
Here’s how Tray answers the questions above.
| Question | Tray.ai's answer |
|---|---|
| Independent audits | SOC 1 Type 2 and SOC 2 Type 2, examined by Schellman & Company, LLC. Reports are available under NDA through the Trust Center.1 |
| Penetration testing | Annual independent penetration tests, the most recent on 23 September 2026, plus a bug bounty program and a HackerOne vulnerability disclosure program.1,2 |
| Encryption | All data is encrypted at rest and in transit, and credentials such as tokens and API keys get additional safeguards.1 |
| Credentials in logs | Workflow authentications are redacted from workflow input logs.3 |
| Acting as the end user | Tools published through Agent Gateway can run with the end user's own credentials.4 |
| SSO and two-factor | Tray enforces SSO and phishing-resistant two-factor authentication.1 |
| Roles and separation | Role-based permissions, with workspaces to separate teams or environments such as development and production.1,5 |
| Support access | Support sees your data only when you grant time-limited access, and every request and action generates an audit event.1 |
| Audit logging | Every action across the platform is logged, and logs can stream to your own systems.1 |
| Log retention | Set retention from 30 days down to 24 hours, or stop storing it.1,3 |
| Masking | Admins can mask step inputs and outputs in logs, and every reveal is audited.6 |
| Hosting regions | US, EU, or APAC.1 |
| Backups and recovery | Multi-region AWS hosting, data backed up at least daily, and disaster recovery tested annually with defined recovery objectives.1 |
| AI data | Customers own AI output and can opt out at any time. Tray never trains on your data, and AI providers don't retain prompts or responses after processing.1 |
| Webhooks | Webhook URLs include a UUID, and Tray recommends requiring a CSRF token or putting the workflow behind API management.3 |
| Contract | A standard DPA and a current sub-processor list. Confirmed incidents trigger investigation and customer notification.1,7 |
| Privacy laws | GDPR, CCPA, and HIPAA, and the EU-US, Swiss-US, and UK-US Data Privacy Frameworks.1 |
Some of these depend on your plan. Regional hosting is standard on Enterprise and an add-on on Team, and log streaming and the shorter retention settings aren’t available on Pro.3,8,9 Check the plan you’re buying covers the controls your review requires.
Running the review quickly
- Request the reports first. The Trust Center holds certifications, audit reports, sub-processor lists, and privacy commitments, with reports available under NDA.1
- Map the answers to your own questionnaire rather than sending a blank one. Most of what it asks is answered on the pages cited here.
- Test your architecture during the trial. Connect a non-production system with a least-privilege account, secure a webhook, set retention, and check what appears in the logs and the audit trail.
For what the logs themselves should record, here’s integration monitoring and audit logs. For governing what AI agents can reach, see governing the MCP servers and AI agents you already run.
Sources
- Tray.ai, Trust: https://tray.ai/trust/
- Tray.ai documentation, Security policies: https://tray.ai/documentation/platform/enterprise-core/security-compliance/security-policies
- Tray.ai documentation, Building securely: https://tray.ai/documentation/platform/enterprise-core/security-compliance/building-securely
- Tray.ai documentation, Agent Gateway overview: https://tray.ai/documentation/platform/artificial-intelligence/agent-gateway/overview
- Tray.ai documentation, Roles: https://tray.ai/documentation/platform/enterprise-core/organisation-management/users/roles
- Tray.ai documentation, Log masking: https://tray.ai/documentation/platform/enterprise-core/security-compliance/log-masking
- Tray.ai documentation, Data protection commitment: https://tray.ai/documentation/platform/enterprise-core/security-compliance/data-protection-commitment
- Tray.ai documentation, Regional hosting: https://tray.ai/documentation/platform/enterprise-core/organisation-management/regional-hosting
- Tray.ai, pricing: https://tray.ai/pricing/
