Docebo Agents — Technical Brief for IT & Security Leadership
Governed, permission-aware AI agents inside your existing security perimeter
Docebo Agents are purpose-built AI systems that execute human-approved actions across Docebo and connected systems. They operate under your existing identity, access, and data-governance controls. No new privilege layer, no elevated service accounts, and no exposure of customer data to model training. Every action is scoped, inherited from source-system permissions, and recorded for review.
Architecture & data sources
Agents are grounded in your governed learning data and explicitly approved knowledge sources, so outputs are derived from authoritative context rather than open-ended generation.
Governed learning data
Courses, lessons, completions, learner activity, and learning plans already resident in your Docebo tenant.
Approved knowledge sources
Scoped Knowledge Collections of third party content and uploaded files that constrain each agent to approved, source-permissioned content.
Identity & access control model
Agents enforce least-privilege by design. Access is inherited from the systems you already govern—there is no parallel permission model to provision, reconcile, or audit separately.
Inherited, least-privilege access
Agents respect the entitlements of each connected system and cannot exceed the authorization already granted to the invoking user or service.
Role-, group-, and branch-scoped
Administrators define which users, groups, and org branches can invoke each agent through existing RBAC constructs.
Pre-authorized action set
Admins declare the exact actions an agent may perform at configuration time; nothing executes outside that approved boundary.
Scope-bound grounding
Knowledge Collections remain within their source permissions and defined scope, preventing lateral data access.
Security, compliance & data handling
Agents inherit the controls, certifications, and data-residency posture of the Docebo platform—no separate security review of a bolt-on AI stack.
No customer data in model training
Your tenant data is never used to train, fine-tune, or improve foundation models.
Permission-aware execution
Agents access only the data and systems the invoking identity is already authorized to use.
Certified control environment
Agents run within Docebo’s certified environment—ISO 27001, SOC 1, SOC 2, GDPR, and CCPA.
Vetted models with guardrails
Docebo integrates pre-trained foundation models from trusted providers with built-in guardrails and content-safety filtering.
Administration & governance
Agents are configured, provisioned, and monitored in AgentHub—Docebo’s no-code control plane for activating, scoping, and observing agent behavior across the tenant.
No-code configuration
Administrators define an agent’s context, action set, triggers, and audience in plain language—no engineering resources required.
Controlled provisioning
Superadmins create custom agents; per-agent access is administered through group and branch entitlements.
Monitoring & iteration
Review agent runs and outcomes, then adjust scope, context, and actions over time.
Auditability & oversight
Every agent operates transparently: actions are pre-authorized, attributable, and reviewable, giving your security and compliance teams the visibility they expect from any automated system.
Attributable actions
Agent activity is tied to the configuring administrator and the entitlements under which it runs.
Human-in-the-loop by design
Administrators approve the action boundary up front and retain the ability to pause, adjust, or retire any agent.
