Roadmap

Phased so the hub is useful as early as possible: “see my apps, launch one, admin can turn access on or off” is the entire MVP bar.

1. MVP
2. Phase 2
3. Phase 3
4. Tenancy automation isn’t phase-gated — it’s revenue-gated
5. The MCP server tracks the REST API automatically
6. What’s explicitly not on this roadmap
7. Where this runs

---

MVP

Enough to replace “someone manually emails a link and flips a flag in a spreadsheet” with a real system.

• User accounts: create, suspend, basic lifecycle.
• Profile — global only. No per-app profile yet.
• Application catalog — list and fetch; writes are admin/internal only.
• Entitlement — grant, revoke, and the on/off toggle. This is the feature the rest of the MVP exists to support.
• Platform Roles only — no app-scoped roles yet; every entitled user is an implicit full member of the app they own.
• Basic Audit Event log for entitlement changes.
• Tenant ships structurally — every Application owner gets one automatically, at tier: shared, the moment they register their first Application. No tier change is reachable yet (that’s Phase 2); the tenant_id column and its Row-Level Security policy exist and are enforced from day one, same reasoning as organization_id below.

Ships: a dashboard-ready API — list a user’s apps, their status, and let an admin flip that status — plus the data model other phases build on without a migration.

Phase 2

• App-scoped Roles & Permissions — apps start defining their own role vocabulary.
• Per-app Profile and Settings, with schema validation against what each Application declares.
• Webhooks — entitlement.*, role.* — so downstream apps can react instead of poll.
• The live introspection endpoint (effective-permissions) — see Trust Model.
• Organizations / seats — team plans, delegated admin via OrganizationMembership.role: org_admin (an org admin manages their own members without needing platform-admin rights), and org-wide Entitlements’ member_scope allowlist/denylist. Pulled forward from a later phase: end users are expected to be both individuals and teams from early on, not teams-later — see Decisions → Organizations.
• Support-run Tenancy tier changes become operationally available — isolated and dedicated_region are real, reachable tiers via the support-gated POST /v1/tenants/{id}/tier-change-requests flow, even though there’s still no customer self-serve path. See Decisions → Tenancy tiers.

Ships: the full per-app customization model, team accounts, and the mechanics apps need to actually trust the hub in production rather than trusting it “eventually, on next login.”

Phase 3

• Richer audit/compliance views — filtering, export, retention policy enforcement; see Compliance & Data Protection for what’s driving this (SOC 2 readiness, GDPR data-subject requests).
• SCIM-style provisioning for larger customers who want their own IdP to push user lifecycle events into the hub automatically.
• Self-service Application registration and review queue for third-party developers — the marketplace phase. See Decisions → App developer/publisher model.

Ships: the features that only matter once there are customers (or outside developers) big enough to need them — deliberately not pulled earlier, since building any of these before there’s a real need tends to guess the shape wrong.

Tenancy automation isn’t phase-gated — it’s revenue-gated

Self-service, customer-triggered Tenancy tier changes (no support ticket, no human running the migration) deliberately don’t have a phase number above. It’s not a feature-scope decision like the rest of this page — it’s a decision to accept a real amount of migration risk (a failed cutover with no human checking each step) in exchange for support time, and that trade only makes sense once tier-change request volume justifies it. Track it against actual demand, not a calendar phase — see Decisions → Tenancy tiers.

The MCP server tracks the REST API automatically

MCP Server also doesn’t have a phase number, for the opposite reason from Tenancy automation above: it needs none, because its tool surface is generated directly from openapi.yaml (see MCP Server → Tool surface is generated, not hand-authored). Whatever of the REST API is live at MVP is agent-callable at MVP; whatever Phase 2/3 adds becomes agent-callable the same day, with no separate MCP work item to schedule. The only real build step — adding operationId to new operations as they’re written — is already folded into writing the endpoint, not a follow-up task.

What’s explicitly not on this roadmap

• Any end-user interface. This is an API-first roadmap; a UI project consumes whatever phase is live, on its own timeline. See Getting Started → What’s deliberately not here.
• Payment processing. Orders/Subscriptions are referenced, not owned, by this API — see Decisions → Billing system of record.

Where this runs

These phases are about feature scope, not infrastructure — the MVP and Phase 2/3 all deploy onto the same Deployment Architecture, which has its own, independent phasing (first deployment vs. scale-out) driven by traffic and cost, not by which of these feature phases is live.