Agents & capabilities

Governed teammates, not bots

In Nexa Agora, agents are not faceless automations. They are named members of the organization: role-based, configurable, capability-bounded, and accountable inside the same operating model as the humans they support.

An agent should not feel safe because it sounds confident. It should feel safe because its actual limits are inspectable.
The anatomy

What makes an agent a teammate

The important point is not the number of agents. The important point is that every agent has a defined role and visible boundaries. Trace each part on the registry sheet.

NameThe agent is identifiable.
RoleThe agent has a place in the organization.
ScopeThe agent knows what kind of work it supports.
CapabilitiesThe agent can use only approved abilities and tools.
BoundariesThe agent’s limits are visible before work begins.
LunaREGISTRY SHEET · AI TEAMMATEAvailable
NameLuna
RoleLead writer — drafting and editorial support
ScopeMarketing and content projects
CapabilitiesMemory · Documents · Review
BoundaryCannot publish without approval
The registry

The agent roster

Named, scoped, assigned, and accountable.
SilvestroOrchestrator
TessaProject manager
LunaLead writer
AuroraGTM strategist
LumenDesign expert
ChobinOperations intelligence
BeatriceOnboarding guide
VirgilioTechnical guide
VeraWeb research
KeplerData analyst
CassandraAdversarial research
DedalusSolution architect
ArchieDeliverable specialist
ArgoReviewer
AxiomPrompt engineer
AureliaEU regulation
Registry sheet
Luna Lead Writer

Drafting and editorial support — briefs, articles, newsletters — always as drafts that move through human review.

CapabilitiesMemoryDocumentsReview
BoundaryCannot publish without approval
StatusAvailable

Teams can also create their own specialist agents for the workflows they run.

Guided product momentOpen a profile · inspect role · view capabilities · check boundaries.
Trust models

Three classes of agents

System agents

The first-party teammates the workspace ships with — the roster above. Defined and maintained by Nexa, they hold the platform's core roles and cannot be modified or deleted.

Internal agents

Agents your team creates — and can delete — for its own workflows: granted only capabilities the platform allows, tuned to your specific needs.

External agents

Outside intelligence admitted through the Nexa Bridge — assistants connected through MCP, your own models and devices — always under a human owner, inheriting that human's permission boundary.

External intelligence does not bypass governance. It enters the workspace under a human's authority.
EXTERNALINTERNALSYSTEMNBenters under a human's authority
The capability gate

Prompt is not permission

A prompt can describe what an agent is supposed to do — tone, role, expertise, behavior. But it does not grant authority. What an agent can actually do is the intersection of two layers.

CONFIGURATIONwhat the agent is designed to doPERMISSIONSwhat the agent is allowed to doREALCAPABILITYPrompt defines intention. Permissions define authority.
The compiler suggests.
The platform authorizes.
The agent executes.
The audit records.
Guided product momentPrompt vs permission — blocked until authorized.
Guided product momentA missing capability, surfaced with next options.
The proof surface

Agents Studio turns agent creation into an organizational act

Creating an agent is not just writing a prompt. It means deciding what kind of teammate the organization needs, what work that teammate can support, which model it uses, what memory and tools it may access — and where its authority must stop.

Guided product momentCreate a specialist agent · define role · choose model · grant capability · set boundary.
Trust badges

Limits should be readable

Capabilities should not be understandable only to engineers. Trust badges make agent limits readable at a glance — what is enabled, what is blocked, and what requires approval.

Luna — capability badges
Memory accessWeb accessTool accessDocument write — approval requiredEU-resident modelPublishing — blockedExternal systems — not configured
enabledapproval requiredblockedmodel residency
Human authority

Autonomy is granted in measured doses

Nexa Agora avoids two weak extremes: agents too passive to help with real work, and agents acting invisibly beyond human control.

AssistProposeExecute within scopeRequires approval
Agents can propose. Humans can approve.
The system records what happened.

Agents work where the work lives

Useful AI work is rarely a single answer. It involves context, coordination, memory, review, and follow-through — so agents participate across the whole workspace, not only in chat.

Agora discussionsProjectsTask boardsPlaybooksDesign StudioCompany BrainOperationsPending Actions

Design your first governed AI teammate

Start with one role your team already needs: research, review, coordination, onboarding, operations, or delivery support.