How Trinity scopes and protects your data
Trinity reads only what your organization connects and each project selects, keeps every organization in its own tenant and does not train AI models on your data.
Security status at a glance
- SOC 2
- Program underway with Vanta, which continuously monitors our controls. No audit report has been issued yet. See the Trust Center.
- Hosting
- Google Cloud in the United States (us-central1, Iowa). The public website runs on Cloudflare.
- AI model training
- Trinity does not train AI models on customer data.
- Security questionnaires
- Completed on request. Email hi@usetrinity.ai.
- Data processing agreement
- Available on request from hi@usetrinity.ai.
- Privacy law
- UK and EU GDPR and the CCPA, as set out in the Privacy Policy.
- Sub-processors
- Listed on the Trust Center and in the Privacy Policy.
- Vulnerability reports
- Email hi@usetrinity.ai with "security" in the subject. We aim to acknowledge within one business day.
How data moves through Trinity
Data comes in from the tools you connect, scoped by each project, and is held in your organization's own tenant. AI providers get the context a request needs, and every answer cites its source.
Sources
Tools your organization connects
Slack
Granola
Google Calendar
Linear
Jira
GitHub
Notion
Claude Code, Codex, Cursor: Coding sessions on paired machinesScope
Each project selects
- Channels
- Repositories
- Issues
- Pages
Pause or disconnect a tool and new capture stops.
Trinity
Your organization's own tenant
- Google Cloud in us-central1 (Iowa)
- TLS in transit, encrypted at rest
- Connector credentials: AES-256-GCM, a key per provider
- Every answer scoped to one project
- No AI training on your data
AI providers
The context a request needs
- Sent through business APIs
- Bring your own model keys Enterprise
Answers
Every answer cites its source
- Slack message
- Pull request
- Issue
- Doc
- Meeting note
Never captured
- Slack direct and group messages
- Channels no project selected
- Anyone's email inbox
Controls you configure
What your organization connects, selects and decides, and the rules that hold whatever it chooses.
Trinity reads only what you select
Access is decided twice: the organization connects a tool, then each project selects what Trinity may read from it. Coding sessions are captured only on machines paired with Trinity. Nothing else is captured.
Connected by the organization
The organization connects Slack, GitHub, Linear, Jira, Notion, Granola or Google Calendar. Nothing is read from a tool that is not connected.
Selected by each project
Each project selects the Slack channels, repositories, issues and pages it covers, and sees only those.
Never captured
Trinity never captures Slack direct or group messages, channels no project selected or anyone's email inbox.
Pause or disconnect
Pausing or disconnecting a tool stops all new capture. What was already captured stays until it is deleted.
Identity and access
People sign in with accounts they already have, and organization owners decide who is in and what role each person holds.
No passwords stored
Sign in with Google or GitHub. Trinity stores no passwords.
Roles
Organization owners invite members and manage roles. Organizations have owners and members, and projects have admins and members.
SSO, SAML and SCIM
EnterpriseEnterprise adds SSO, SAML and SCIM provisioning.
Retention and deletion
Captured data stays until it is deleted. An owner can delete the whole organization, and Enterprise adds custom retention windows.
Delete the organization
An organization owner can delete the organization from Settings. That removes connector credentials, sessions and captured content, including AI workspace data, and cancels billing.
Retention controls
EnterpriseEnterprise adds audit logs and custom data retention windows.
Coding agents you can account for
Sessions from Claude Code, Codex and Cursor are captured on paired machines, linked to the task they served and visible with their usage and cost.
Paired machines
Coding sessions are captured on machines paired with Trinity, and any paired machine can be revoked.
Linked to the task
Each session is linked to the task it served, so the work an agent did can be traced.
Writes carry the person's name
An agent on a paired machine writes back through Trinity as the person who paired it: starting a task, reporting progress, raising a blocker, linking a pull request or finishing a task. Every write carries that person's name.
Done follows the merge
A task is not marked done while an open pull request implements it. Completion follows the merge.
Usage and cost
Token usage, cost and sessions are visible across the project.
Run it where your policies require
Trinity's cloud runs on Google Cloud in the United States. On Enterprise, Trinity can also run in a private cloud or self-hosted.
Trinity's cloud
Hosted on Google Cloud in us-central1 (Iowa). The public website runs on Cloudflare.
Private cloud or self-hosted
EnterpriseEnterprise can be deployed in a private cloud or self-hosted.
Uptime commitment
EnterpriseEnterprise agreements carry a contractual uptime commitment, with the numbers set in the agreement.
Controls we operate
What Trinity runs on your behalf.
Every organization is its own tenant
Data is kept apart at two boundaries: each organization, and each project inside it.
Organization boundary
Each organization is a separate tenant. Nothing crosses from one organization to another.
Project scoping
An organization holds multiple projects. Every answer and automation runs inside one project and sees only what that project selected.
Org-wide view on Enterprise
EnterpriseEnterprise adds unlimited workspaces with org-wide visibility, still inside your organization's own tenant.
Encrypted in transit and at rest
Data moves over TLS and is encrypted at rest by Google Cloud. The credentials Trinity holds for your connected tools get a further layer of encryption.
In transit
Data in transit is protected with TLS.
At rest
Google Cloud encrypts stored data at rest.
Connector credentials
Credentials for connected tools are additionally encrypted with AES-256-GCM, with a separate key for each provider.
Your data does not train AI models
Trinity uses AI providers to answer questions and run automations. It sends them the context a request needs, through their business APIs.
No model training
Trinity does not train AI models on customer data.
Business APIs
Each request sends the context it needs to Trinity's AI providers through their business APIs.
Your own keys
EnterpriseOn Enterprise, bring your own model keys and pin a model per automation.
Answers you can check
Each answer links to the Slack message, pull request, issue, doc or meeting note it draws on, so you can check it.
How we run the service
Two controls keep the service in check: restricted production access and continuous monitoring.
Production access
Production access is restricted to the engineers who run the service, each with unique credentials.
Continuous monitoring
Vanta monitors our controls continuously.
Compliance status and documents
We publish where we stand, not ahead of it. Until a SOC 2 report is issued, we complete security questionnaires, and a data processing agreement is available on request.
SOC 2 in progress
Our SOC 2 program is underway with Vanta. No audit report has been issued yet.
Trust Center
Our controls are published at trust.usetrinity.ai.
Security questionnaires
Until a report is issued, we complete security questionnaires. Email hi@usetrinity.ai.
Data processing agreement
A data processing agreement is available on request.
Privacy law
How Trinity handles personal data under the UK and EU GDPR and the CCPA is set out in the Privacy Policy.
Sub-processors
The current list of providers that handle data for Trinity is published on the Trust Center and in the Privacy Policy.
Vulnerability reports
Found a security issue? Email hi@usetrinity.ai with "security" in the subject. We aim to acknowledge within one business day.
Questions reviewers ask
Anything else? Write to hi@usetrinity.ai.
Not yet. Our SOC 2 program is underway with Vanta. Until a report is issued, we complete security questionnaires on request.
No. Trinity does not train AI models on customer data. It sends the context a request needs to its AI providers through their business APIs.
On Google Cloud in the United States (us-central1, Iowa). Enterprise can also run in a private cloud or self-hosted.
No. Trinity never captures Slack direct or group messages, channels no project selected or anyone's email inbox.
No. Each organization is its own tenant, and inside it every answer is scoped to one project.
An organization owner deletes the organization from Settings, which removes connector credentials, sessions and captured content.
Yes, on request.
Yes, on Enterprise: SSO, SAML and SCIM provisioning.