SECURITY BY ARCHITECTURE

Connect the work. Keep your boundaries.

Tentrico coordinates AI-native engineering without requiring every part of your development environment to collapse into a single trust domain.

Explore the Architecture

This page describes Tentrico’s security architecture and design direction. Specific controls and availability are confirmed during security review.

  1. Tentrico control planeRequirements · Missions · AI coordination
  2. Policy boundaryAuthorization decides what crosses
  3. Your environmentSource · Credentials · Execution · Repositories · Cloud · CI/CD

TRUST BOUNDARIES

Coordinate the system. Don’t collapse the boundaries.

Tentrico connects the objects required to coordinate software delivery while preserving separation between the control plane and the environments where engineering work executes.

Tentrico control plane

Coordinates the work

  • Requirements
  • Missions
  • Decisions
  • Human Tasks
  • Evidence metadata
  • Policies
  • Release metadata
  • Audit events
  • Policy
  • Connector
  • Audit

Your environment

Customer-controlled

  • Source code
  • Credentials
  • Developer workspaces
  • Agent execution
  • Cloud resources
  • Local tooling
  • Runtime systems

Intended architecture. Source code, credentials and execution can remain in your environment; confirm supported configurations during security review.

Your source code doesn’t need to live in our cloud.

Tentrico’s shared control plane can coordinate requirements, missions, decisions, policies and evidence while source code, development credentials and execution remain inside customer-controlled environments.

Shared control planeCoordination state
  • Requirements & missions
  • Decisions & policy results
  • Evidence metadata
Source codeRemains in your repositories and working copiesCredentialsExecution

WHAT CROSSES THE BOUNDARY?

Share coordination. Keep execution controlled.

The control plane receives what it needs to coordinate the work. The assets that do the work can stay where they are.

Large or sensitive evidence artifacts are designed to stay in customer-controlled storage, with only their metadata shared.

Control plane data

  • Requirement records
  • Acceptance criteria
  • Mission state
  • Decision records
  • Human Task state
  • Evidence metadata
  • Release metadata
  • Environment definitions
  • Policies
  • Audit events

Customer-controlled data & execution

  • Source repositories
  • Local working trees
  • Development credentials
  • Raw terminal activity
  • Local agent execution
  • Customer cloud resources
  • Runtime credentials
  • Sensitive environment configuration

Intended data placement in Tentrico’s architecture. Exact storage, retention and residency depend on the supported deployment configuration.

SECRETS

Credentials are capabilities. Not conversation.

Sensitive credentials should be referenced through approved secret stores and connectors rather than casually exposed in prompts, logs or agent conversation.

  1. AgentRequests a capabilityDEPLOY_TO_QA
  2. TentricoResolves the authorized connectorconnector: qa-deploy
  3. Secret referencePoints to a credential, never contains itref: qa/deploy-token
  4. Secure storeHolds the credentialvalue stays here
  5. TargetQA environmentdeploy accepted

What the agent receives

Capability
DEPLOY_TO_QA
Credential
configured
Authorization
granted
Result
success

What it doesn’t

The credential value

The credential value stays in the secure store. Secure-store concepts: OS credential store · Customer secret manager · Workspace secret reference.

Secure-store options are architecture concepts; supported stores and connectors vary by release.

LEAST PRIVILEGE

Give agents capabilities. Not unlimited access.

Instead of holding broad credentials, an agent requests a narrowly defined capability. Each request is evaluated on its own.

Each capability is evaluated against

  • Identity
  • Workspace
  • Environment
  • Policy
  • Risk

Instead of

Agent holds broad production credentials

Prefer

Example capabilities with their environment scope and risk
CapabilityScopeRisk
READ_REPOSITORYLocal · DevLow
CREATE_BRANCHDevLow
RUN_TESTSDev · QALow
READ_QA_LOGSQALow
DEPLOY_QAQAMedium
PROPOSE_PRODUCTION_RELEASEProductionHigh · proposes only

Illustrative capability names describing the architecture, not a published capability list.

POLICY

An agent request isn’t authorization.

Tentrico’s policy engine sits between AI intent and privileged action. Every request is evaluated before anything executes, and the decision belongs to policy, not to the model.

Agent requestTentrico

Policy engine

  • Identity
  • Workspace
  • Environment
  • Capability
  • Risk
  • Evidence
  • Approval
  • Credential reference
  • Allow
  • Require human
  • Deny

Action request

Deploy REL-2.14

Requested by
Deployment Agent
Target
Production
Workspace
Authentication Platform
Environment
PRODUCTION
Capability
DEPLOY_RELEASE

Policy evaluation

  • IdentityVerified
  • ReleaseREL-2.14
  • EvidenceComplete
  • SecurityPassed
  • EnvironmentPRODUCTION
  • PolicyPOL-PROD-04
  • Human approvalRequired
ResultWaiting for approvalPaused. HT-0094 is created for the Release Manager.

Illustrative product preview · sample data

Example policies and identifiers are sample data. Policy definitions are configured per organization and environment.

HUMAN AUTHORITY

Autonomy stops where judgment begins.

When policy requires human authority, Tentrico pauses the action and creates a structured Human Task for the appropriate person.

HT-0094Human Task

Production release approval

Waiting for human
Release
REL-2.14
Environment
Production
Requested by
Deployment Agent
Policy
POL-PROD-04
Evidence
Complete
Security
Passed
Assigned
Release Manager

The deployment is paused at the boundary until the Release Manager decides.

Illustrative product preview · sample data. These buttons only change this preview; nothing is deployed.

  1. RequestAgent asks to deploy
  2. EvaluatePolicy checks identity, evidence, environment
  3. Human requiredPOL-PROD-04 asks for a person
  4. ApprovalWaiting for a person
  5. Authorized actionNot yet

ENVIRONMENT-AWARE GOVERNANCE

Production isn’t just another destination.

Tentrico can apply different policy, evidence and approval expectations depending on where an action is headed.

  1. Local
    • Low risk
    • Developer-controlled
  2. Dev
    • Automated
    • Standard checks
  3. QA
    • Verification required
  4. UAT
    • Human acceptance
  5. Production
    • Evidence
    • Policy
    • Authorization
    • Human approval where required

A conceptual policy model. Each organization defines its own environment policies; these are not product defaults.

REL-2.14

Production gate

  • Evidence passed
  • Security passed
  • Approval passed
  • Policy passed

Authorized

Production

CONNECTORS

One governed path to external systems.

Agents don’t reach external systems directly. Requests travel through Tentrico and policy to a connector, the one place that holds the integration.

Tool protocols such as MCP are designed to sit behind the same policy layer, not around it.

  1. Agents
  2. Tentrico
  3. Policy
  4. Connectors
  • Git provider
  • CI/CD
  • Cloud
  • Observability
  • Security tools
  • Identity

Categories of system, not a list of supported integrations or partners.

TRACEABILITY

Know what happened. Know why it happened.

Every important action connects back to the requirement that asked for it, the policy that governed it and the person who authorized it.

Audit trail

Deploy REL-2.14 to Production

  1. RequirementREQ-AUTH-014Enterprise SSO
  2. MissionMIS-0182Implement SAML login
  3. Agent runRUN-0087Deployment Agent
  4. Action requestedDeploy REL-2.14Target: Production
  5. PolicyPOL-PROD-04Human approval required
  6. Human TaskHT-0094Production release approval
  7. Approved byRelease ManagerIdentity verified
  8. DeploymentDEP-0042Completed
  9. EvidenceEV-0218Post-deploy verification
  10. StatusVERIFIEDRequirement satisfied in Production

Events · REL-2.14

  1. deployment.requested
  2. policy.evaluated
  3. human_task.created
  4. human_task.approved
  5. deployment.authorized
  6. deployment.completed
  7. verification.passed

Security-sensitive actions produce traceable system state.

Illustrative product preview · sample data. Event names are illustrative.

MODEL INDEPENDENCE

Security doesn’t depend on the model behaving perfectly.

AI models can propose work. Tentrico’s authorization, policy and connector layers determine whether sensitive actions are allowed to execute.

Model / agentProposes
Policy boundaryAuthorizes
Authorized actionExecutes through a connector

The agent proposes. Policy holds the key.

DEPLOYMENT

Choose the boundary that fits your organization.

The same security model at every scale. What changes is where the control plane runs and who operates it.

  • Personal / Solo

    Local control plane, local execution

    • Tentrico Desktop
    • Local state
    • Local agents
    • Local execution
    • External AI providers as configured by the user

  • Team

    Shared control plane, developer-controlled execution

    • Tentrico Desktop and runners
    • Shared Tentrico control plane
    • Shared requirements, missions, decisions and evidence
    • Developer- and customer-controlled execution

  • Enterprise

    Planned

    Organization-wide control plane, customer-controlled execution

    • Managed or supported self-hosted control plane
    • Enterprise identity
    • Policy
    • Audit
    • Customer-controlled execution
Tentrico managedPlanned
  1. Tentrico-hosted control plane
  2. Secure, authenticated outbound connectionDesktop / runner
  3. Customer environmentSource · Agents · Tools · Infrastructure

Connections are designed to start from your desktops and runners. Tentrico doesn’t need arbitrary inbound access to developer machines.

Self-hostedPlanned
  1. Tentrico control planeIn your AWS, Azure, GCP or other supported infrastructure
  2. Customer identity
  3. Customer runners / desktops
  4. Repositories · CI/CD · Cloud

Self-hosting is a planned option. Platform support is confirmed during architecture review.

Enterprise security boundary
  1. Outside providersAI providers · External services
  2. Controlled connectionTentrico control plane
  3. Policy · Identity · Audit
  4. Controlled connectionCustomer execution environmentSource · Agents · Cloud · CI/CD · Runtime

Connections are explicit. Trust domains stay visible.

Managed and self-hosted control planes, enterprise identity and organization-wide policy are planned capabilities, not a statement of current general availability.

SECURITY PRINCIPLES

Where authority lives.

  • Boundaries are explicit

    Sensitive systems stay separated by clear trust boundaries.

  • Authorization is independent of AI

    Models propose. Policy authorizes.

  • Humans retain authority

    High-risk decisions can require explicit human approval.

  • Secrets are references

    Sensitive credentials should remain in approved secret stores.

  • Production is governed

    Higher-risk environments can require stronger controls.

  • Actions are traceable

    Important actions connect back to identity, intent, policy and evidence.

Bring your security questions.

Review source access, credential handling, agent-provider policies, retention and deployment boundaries with us. No certification or compliance attestation is claimed on this page.

  1. Policy
  2. Authorization
  3. Connector
  4. Your systems

CONTROL WITHOUT SURRENDERING CONTROL

Your code. Your infrastructure. Your policies. One governed system.

Connect AI-native engineering without giving up the boundaries that protect it.