Skip to content

Product Information Architecture

Target architecture — not the current operating contract

This navigation model includes future products. For current console navigation, see the User guide and Administrator guide.

The current console groups work under Access, Observe, Evaluate, Build and Manage, with Chat for audience members; Observe provides Requests, Traces, Sessions and Usage, and there is no current Runs destination. The table below retains a possible future Run Explorer only as target architecture.

Mission Control is one role-aware application over a shared estate. Navigation groups work by user intent while resource pages and cross-links preserve one identity, lifecycle, authority, and evidence model across modules.

Group Destination Primary purpose
— Overview Role-aware estate summary, attention queue, health, approvals, releases, cost, outcomes, and recent activity
Build Agents Design, import, test, evaluate, version, release, and operate Agents and Agent Teams
Build Automations Design, test, release, and operate durable Workflows
Build Applications Compose, preview, publish, embed, and operate AI Applications
Build Retrieval Register sources and manage, evaluate, and operate Retrieval Collections
Connect Connectors Register MCP Servers and classify, certify, grant, version, and operate individual MCP tools
Connect Models Register providers and deployments; manage Virtual Models, routing, access, budgets, guardrails, evaluation, and health
Control Policies Author, simulate, review, publish, bind, explain, and analyze policy
Control Approvals Process Governance Approvals and Human Tasks while preserving their distinct contracts
Control Identity and Access Configure SSO mappings, human roles, workload identities, service credentials, grants, delegation, and environment scope
Operate Estate Inventory resources, owners, versions, lifecycle, dependencies, access, policy, health, and incidents
Operate Runs Explore Run and Trace structure, decisions, waits, retries, calls, costs, failures, evidence, and Outcomes
Operate Evaluations Manage suites, release comparisons, production cohorts, safety, quality, cost, and business Outcomes
Operate Cost and Capacity Attribute and control model, Connector, runtime, Retrieval, telemetry, evidence, and storage usage
Operate Incidents Investigate degradation through affected resources, versions, Traces, decisions, changes, corrections, and Outcomes
Operate Evidence Search, reconstruct, inspect, and export controlled operating evidence
Platform Environments Define lifecycle stages, environment isolation, bindings, network posture, and operating controls
Platform Managed releases Review future Agent Builder/native releases, attestations, health, promotion, suspension, and rollback; current external executions remain visible as Traces in Observability
Platform Settings Configure enabled modules, gateway behavior, retention, redaction, notifications, integrations, and customer preferences

Navigation labels are product terms, not organizational names. A resource has one canonical page even when it is reachable from several work profiles.

Shared resource pages

Every first-class resource page uses the same top-level anatomy:

Resource header
  Name + type + environment + lifecycle + version + owners + health

Overview | Versions | Access | Policy | Activity | Evidence
Tab Contract
Overview Purpose, current state, owners, environment, health, dependencies, key metrics, active version, recent attention, and module-specific summary
Versions Mission-Control-managed definitions/releases, comparison, review, approvals, promotion, attestation, active bindings, rollback, suspension, and retirement
Access Human and workload grants, requested capabilities, groups, delegated authority, environment scope, and effective access
Policy Bindings, active policy versions, simulated and actual decisions, approval routes, obligations, and explanations
Activity Attributed lifecycle and administrative changes plus linked Runs, tasks, evaluations, incidents, and operational events
Evidence Durable resource-specific Evidence Events, corrections, content-capture status, retention, exports, and business Outcomes

The resource header makes high-consequence state visible before any action. Lifecycle, environment, version, owner, health, risk, and governance state never depend on color alone. Primary actions are permissioned, version-aware, and separated from destructive or state-changing actions.

Cross-resource navigation

Run and Trace context provides the primary cross-resource spine. A Run can link directly to:

  • Its initiating user, workload, AI Application, Agent, Workflow, trigger, and business reference.
  • Every participating Agent or Agent Team, grants, keys/workload identity, Runs, and any applicable managed release/runtime.
  • Virtual Model, routing policy, resolved model deployment, provider call, guardrail, token usage, and cost.
  • MCP Server, Connector, tool-level class, certified version, policy decision, Governance Approval, invocation, result reference, and Outcome.
  • Retrieval Collection, source, Data Connector, citation, freshness state, and access evaluation.
  • Workflow step, Human Task, branch, timer, retry, exception, compensation, and durable state.
  • Evaluation Suite and result, incident, correction, evidence record, and export.

Links preserve the original environment and historical version context. Opening a resource from an old Run defaults to the version used in that Run and visibly distinguishes it from the currently active version.

Tables, timelines, Trace trees, decision cards, approval cards, incidents, and evidence records expose stable identifiers with copy actions. Search supports display names and identifiers, but navigation never depends on mutable names.

Module availability

The Control Kernel, core Control and Operate experiences, Connectors, Models, Identity and Access, Environments, Deployments, and Settings are mandatory. Composable authoring modules can be unavailable by customer configuration, environment, license, or role.

  • Hide a module when it is not enabled for the customer and no visible resource or action depends on it.
  • Show a disabled navigation or capability with an explanation when the user needs to understand a dependency, request enablement, inspect an existing resource, or complete a journey.
  • Show a permission-denied state, not an absent module, when the module exists but the user lacks access.
  • Preserve historical resource, Run, and evidence access according to policy after a module is disabled; do not break evidence links.
  • Never expose a clickable control that fails only after the user enters a disabled or unauthorized flow.

Responsive behavior

Mission Control is desktop-first for dense operational work and responsive for review, approval, investigation, and monitoring.

  • Desktop uses a persistent grouped sidebar, environment selector, global search or command entry, attention center, and a flexible main workspace.
  • Compact desktop collapses groups to icons with labels on focus or expansion while preserving the current location and environment.
  • Tablet uses an overlay navigation panel, sticky resource header, horizontally scrollable secondary tabs, and column-prioritized tables.
  • Mobile uses a top application bar and navigation drawer. High-value summaries, approvals, tasks, incident status, and evidence details reflow into cards; dense designers and Trace comparisons use a supported read-only or focused subset.
  • Tables preserve semantic headers, offer column priority and detail drawers, and never require horizontal scrolling for the primary identity and state.
  • Timelines and Trace trees keep causal nesting with progressive disclosure rather than flattening away parent-child context.

Responsive layouts preserve the same information and authority hierarchy. They do not remove policy reasons, version context, state labels, or confirmation details to fit smaller screens.

Current governance mapping

The current source release uses Organizations, Teams, real user/service-account virtual keys and separate human management roles. Agent is managed application content; deployments own secretless execution access and publisher funding. Observe uses Requests, Traces, Sessions and Usage with separate metadata/content/export permissions. Historical identity terms in this target proposal are not the operating contract. See Current release.