Explicit authority
Agents act through allowlisted actions and permission boundaries. Connecting a system does not grant unlimited power.
Governed execution
Governed AI execution for real company work
Relsor coordinates AI work inside explicit permission, approval, tenant-isolation, and evidence boundaries — so agents act with authority you can see and control.
Principles
Security principles built into how Relsor operates
Constrained authority, human control, and inspectable results — built into Relsor’s operating model.
Human control
Consequential and outbound actions can require human approval. Rejection blocks execution.
Tenant-scoped work
Supported product workflows enforce tenant boundaries so company work stays inside the authenticated organization.
Inspectable execution
Jobs, approvals, artifacts, and orchestration lineage remain visible where supported — not hidden behind opaque autonomy.
Evidence over assumption
Results return with inspectable evidence in Mission Control so operators can review what ran and why.
Honest capability boundaries
Integrations carry clear maturity labels. Foundation or connectable paths are never presented as equally live.
Execution model
Governed execution — not unlimited agent authority
Owner intent moves through the AI CEO and specialists on allowlisted tool paths. Approval gates interrupt the graph when required. Evidence returns to Mission Control.
- 1Step
Owner intent
A person states the outcome
- 2Step
AI CEO
Coordinates and delegates
- 3Step
Specialist agent
Owns the domain task
- 4Step
Governed tool path
Allowlisted actions only
- 5Step
Approval when required
Graph can pause
- 6Step
Evidence returned
Visible in Mission Control
Connecting an integration does not automatically authorize every agent or every action. Available work depends on permissions, assignment, and the maturity of each connection path.
Control
Permissions and approvals
Humans keep authority over consequential actions. Agents do not receive an open-ended mandate.
Permission boundaries
Agents operate inside explicit permissions. Available actions depend on role, assignment, and the connected integration — not on an open-ended mandate.
Approval-required actions
When an action requires approval, the workflow waits for a human decision. Supported orchestration pauses while approval is pending.
Denial and blocking
Rejected approvals do not execute. Policy and budget gates fail closed when checks cannot be satisfied.
Budget boundaries
Available when configuredBudget and spend controls apply when configured — including gated paid-media foundations. Silent unlimited spend is not the product model.
Isolation
Tenant boundaries
Supported product workflows enforce tenant boundaries. Authenticated sessions carry tenant and role claims; execution and stored work stay scoped to that organization. Cross-tenant access is denied in the product model.
These boundaries apply across supported governed workflows.
Visibility
Audit, evidence, and lineage
Operators can inspect what ran — without exposing hidden chain-of-thought or internal implementation detail.
Jobs and approvals
Operators can inspect job status and approval decisions for work that ran through the governed path.
Artifacts and citations
Where workflows produce artifacts or citations, those outputs remain inspectable alongside the job.
Handoffs and lineage
Multi-step runs can expose handoff references and orchestration lineage so you can follow how work moved between agents.
Mission Control
Mission Control surfaces delegated work, waiting approvals, and evidence without requiring you to guess what the AI did.
Boundaries
Data and provider boundaries
Credentials stay out of handoffs. Provider access depends on what you connect and grant.
No credential transfer in handoffs
Agent handoffs carry bounded context and references — not credentials or secret material.
Bounded downstream context
Downstream agents receive the context required for the assigned task, not an unrestricted dump of tenant secrets.
Provider access is configured
Available when configuredExternal provider access depends on connected integrations and the permissions granted for that connection.
Governed external actions
Supported workflowsExternal actions remain on governed tool paths where supported. Integration maturity labels describe what is live versus foundation-ready.
Secrets
Credentials and secrets handling
Integration credentials are encrypted at rest. Secrets are redacted from logs and client responses. Access uses authenticated sessions with role and tenant claims.
Privacy, legal, and procurement requirements are reviewed during enterprise planning.
Discuss your security requirements
Organizations with specific security, architecture, or procurement questions can review them with Relsor during enterprise planning.