Website and deployment inquiries
Deployment inquiries submitted through this site are stored in Supabase. Worktree does not train models on prospect or customer data.
Product security
Useful agents need access to real business systems. Worktree designs that access around a specific workflow: its required context, permitted actions, approval points, exception paths, operating evidence, and removal.
Security begins with the job
The control model follows the work. Before a workflow receives access, the team should be able to explain its purpose, scope, authority, review path, and end state without relying on vague claims about the agent.
The approved request, event, or schedule that opens the workflow.
The systems and information required for the assigned job.
The read, prepare, update, or submit actions included in scope.
The decisions and consequential actions reserved for an authorized person.
The exception path that stops improvisation and routes the case.
The credential, connection, and retained-data steps at the end of access.
Five control boundaries
Security is not one toggle around the model. The workflow combines access, credentials, authority, human review, and removal into a control design that can be discussed before launch and revisited when the work changes.
The workflow identifies the applications, records, documents, and communication surfaces the agent needs. Connection to one system does not make every part of that system relevant to the role.
Credentials and delegated connections are configured for the agreed workflow. Their ownership, use, and removal should be understandable before production access is granted.
The deployment separates what the agent may read, prepare, update, or submit from the actions that require another person to authorize them.
Known approval points and exception paths keep consequential decisions with the people who own the business policy and outcome.
Delegated access can be disconnected. Retained customer data is deleted following a verified deletion request under Worktree's stated data-handling policy.
Sensitive context
When this control is appropriate to the deployment, sensitive context can be made available for a single authorized agent execution rather than left broadly available to the workflow. The exact design depends on the system, data, and action involved.
Review and recovery
Relevant evidence makes it possible to review an accepted result, trace an exception, discuss a material change, and decide what needs attention next. The exact evidence retained is defined with the deployment rather than implied as universal surveillance.
See how Worktree manages a launched workflowData handling
Deployment inquiries submitted through this site are stored in Supabase. Worktree does not train models on prospect or customer data.
Customer workflow data may be processed with OpenAI when that processing is part of the customer workflow. The relevant systems and handling should be discussed for the proposed deployment.
Information is retained indefinitely by default and deleted following every verified deletion request. Delegated workflow access can also be disconnected.
Worktree does not claim certifications, compliance programs, or security guarantees that are not documented here. A deployment review examines the actual workflow, systems, data, and authority involved.
Controls in practice
Map systems, actions, approvals, test cases, and acceptance criteria before launch.
Explore AI implementationReview exceptions and material change against the agreed deployment standard.
Explore managed AISecurity review
Worktree can help identify the access, authority, approval, data-handling, and removal questions the deployment must resolve. For a security question or verified deletion request, contact hello@orthg.nl.
Last updated August 21, 2026.