What actually slows a team down after choosing an identity provider: the login page, or the daily work of changing it? Most comparisons of Keycloak, Ory, and Zitadel stop at feature checklists. That misses the point. The real difference shows up on a Tuesday afternoon when someone needs to add a second factor, wire up a new tenant, or debug why a password reset silently failed. Each platform asks you to think in a different shape: realms, services, or events. Those shapes become your workflow, and your workflow becomes your velocity. (Keyboard foam mod sound differences actually changes)
This piece walks through how each tool wants you to work, where teams trip up, and what to watch as 2027 approaches.
Why the workflow model matters more than the feature list
All three projects speak OpenID Connect and OAuth 2.0, the standards that underpin most modern sign-in systems. If you want the formal groundwork, the OpenID Foundation’s overview of the OpenID Connect protocol explains what these specifications actually guarantee. Beyond that shared base, the tools diverge fast. (Use claude writing practical workflow that actually works)
A useful way to frame it: Keycloak is a configured server, Ory is a composed toolkit, and Zitadel is an event-driven platform. You can swap features between them on a spreadsheet. You cannot swap the mental model without retraining everyone who touches identity.
Keycloak: the realm-centric admin console

Keycloak is an Apache-2.0 identity and access management server, originally the upstream of Red Hat’s SSO product and accepted as a CNCF incubating project in April 2023. Its organizing concept is the realm: an isolated namespace holding users, clients, roles, and settings.
How a typical change flows
- Log into the admin console and pick the realm.
- Open Authentication and inspect the ordered list of executions for the browser flow.
- Add or reorder an execution, such as an OTP form or a conditional subflow.
- Bind the flow and test it in a staging realm first.
That execution list is powerful and slightly unforgiving. Executions run top to bottom, and a misplaced requirement flag can lock users out of the very console you need to fix it. The practical habit that saves teams: always keep a break-glass admin account in a separate realm, and export realm configuration before editing flows.
Where Keycloak’s workflow shines
Keycloak’s authorization is built on UMA 2.0 and its own Authorization Services, where resources, scopes, policies, and permission evaluation are configured per client. Policy authoring lives inside Keycloak rather than in a separate engine. For teams that want one system to own both authentication and fine-grained authorization, that is a genuine workflow simplification. The Keycloak Authorization Services guide walks through the resource, scope, and policy model in detail. (RFC 6749: The OAuth 2.0 Authorization Framework)
Keycloak also carries the broadest built-in protocol and federation surface of the three, including SAML, LDAP and Active Directory federation, and Kerberos. If your organization still runs on directory servers, Keycloak meets you where you are.
Common mistakes
- Treating realms as environments. Realms are tenants, not dev/staging/prod. Teams that conflate the two end up with configuration drift nobody can trace.
- Hand-editing themes without version control. Theme changes are code; treat them that way.
- Skipping realm exports. A clean JSON export is your rollback plan.
Ory: services you compose yourself
Ory takes the opposite bet. Instead of one monolith, it ships separate open-source services: Ory Kratos for identity, registration, and login; Ory Hydra as the OAuth2 and OIDC provider; Ory Keto for relationship-based, Zanzibar-style authorization; and Ory Oathkeeper as an identity and access proxy. The Ory documentation’s component overview is the best place to see how the pieces fit.
The self-service flow concept
Kratos models login, registration, recovery, and verification as API-driven self-service flows. Each flow has an ID and a set of UI nodes, and your frontend renders them. The Kratos self-service flows documentation explains the flow ID and node structure that makes this work.
That decoupling is the headline workflow difference. Keycloak renders login pages for you through themes. Ory hands your application a data structure and says: build the UI you want. For product teams with strong frontend skills, this is freedom. For teams expecting a working login screen out of the box, it is a project.
What the composed model costs
You become the integrator. A simple email-and-password login might touch Kratos for identity and Hydra for tokens. Add Keto and you have a third service to operate, monitor, and upgrade on its own cadence. Each service is small and focused, but the operational surface multiplies.
Practical advice from teams that run this model: standardize on one deployment pattern early, keep service versions pinned together, and write integration tests that exercise the full login journey rather than each service in isolation. The failure mode is subtle: every service is healthy, yet the flow breaks because one node schema changed.
Zitadel: organizations, projects, and events
Zitadel is an Apache-2.0 identity platform built API-first around organizations, projects, and applications, with multi-tenancy as a first-class concept. Management happens primarily through gRPC and REST APIs and the console. The Zitadel documentation lays out these concepts clearly.
Multi-tenancy as the default
If your product serves many customer organizations, Zitadel’s shape feels natural. An organization maps to a tenant, projects group your applications, and roles attach at the project level. You are not bolting multi-tenancy on with a realm-per-customer pattern; it is the native unit of work.
Actions and event sourcing
Zitadel supports actions, which are JavaScript hooks that customize login and registration behavior, plus flows for orchestrating those steps. Underneath, the architecture is event-sourced: state is derived from an event log rather than mutated in place. That has a real workflow consequence. You can inspect what happened, when, and in what order, which makes audits and debugging considerably calmer than guessing at current-state rows.
The trade-off is conceptual overhead. Event sourcing rewards teams who understand it and confuses teams who do not. Budget time for the mental model before your first production deployment.
Side-by-side: how the work actually differs
| Workflow task | Keycloak | Ory | Zitadel |
|---|---|---|---|
| Custom login UI | Theme overrides on server-rendered pages | Build your own UI against flow nodes | Console branding plus actions and flows |
| Adding a second factor | Reorder executions in the browser flow | Configure the method in Kratos and render it | Configure in console, extend with actions |
| Multi-tenancy | One realm per tenant | Model tenants in your own app layer | Organizations as native tenants |
| Authorization model | Built-in Authorization Services on UMA 2.0 | Keto, relationship-based | Roles on projects, API-enforced |
| Configuration as code | Realm JSON export and import | Service config files and APIs | APIs, with event history as the record |
Read that table as a map of where your team’s hours will go. None of the rows is objectively better. Each is a bet about which skills you already have.
Migration and day-two operations
Day one is setup. Day two is where identity projects quietly succeed or fail.
Upgrades. Keycloak ships frequent releases with occasional breaking changes in themes and APIs. Ory’s separate services upgrade independently, which is flexible but demands a compatibility matrix. Zitadel’s event-sourced core means schema changes deserve a rehearsal in a staging environment.
Observability. With Ory, you must correlate logs across services yourself. With Keycloak, one server log covers most of the story. With Zitadel, the event stream becomes your primary debugging tool once you learn to query it.
Secrets and keys. All three rotate signing keys, and all three will punish an unplanned rotation. Put key rotation on a calendar with a named owner.
What to watch heading into 2027
Be careful with forecasts here. Claims about which project will lead on passkeys, verifiable credentials, or AI-agent identity in 2027 are projections, not facts. As of early 2025, all three were adding WebAuthn and passkey support, and feature parity shifts constantly. Verify current capability matrices at decision time rather than trusting any comparison written months earlier.
Two structural trends are safer to plan around. First, licensing and governance have moved before: Keycloak’s licensing history shifted over the years, Ory’s core is open source with some commercial components, and Zitadel pairs an open-source core with a commercial cloud. Check each project’s LICENSE file at publication time, because boundaries per component can differ. Second, API-first designs age better in a world of automated provisioning, so Ory and Zitadel’s shape may suit platform teams building internal developer platforms. Keycloak’s console-first shape suits teams that want humans, not pipelines, making changes.
If your identity work overlaps with broader tooling decisions, the same “workflow first” lens applies elsewhere. Teams comparing note-taking setups get value from this breakdown of Obsidian vs Logseq local-first notes compared by workflow, and anyone standardizing on an AI assistant will find this guide to Claude alternatives by workflow and price useful context. The pattern is consistent: the tool that fits your daily habits beats the one with the longer feature list.
A practical way to choose
Answer four questions honestly.
- Does your team have frontend engineers who want to own the login experience? If yes, Ory’s node-based flows fit. If no, Keycloak’s themes are kinder.
- Is multi-tenancy a core product requirement? Zitadel’s organizations model saves months of custom work.
- Do you need built-in fine-grained authorization? Keycloak’s Authorization Services keep policy in one place.
- How much operational complexity can you staff? One server is easier than four services, and event sourcing needs a learning budget.
Then run a two-week spike. Build the same login, password reset, and second-factor flow in your top two candidates. The winner is usually obvious by day five, and it is rarely the one with the prettier marketing page.
Conclusion
Keycloak, Ory, and Zitadel all do identity well. They differ in what they ask of you afterward. Keycloak asks you to think in realms and ordered executions, and rewards that with breadth and a single console. Ory asks you to compose services and build your own frontend, and rewards that with flexibility and clean separation. Zitadel asks you to think in organizations and events, and rewards that with native multi-tenancy and an auditable history. Pick the workflow your team can sustain on a slow Tuesday, not the one that looks best in a demo.
FAQ
Is Keycloak harder to customize than Ory?
Not harder, different. Keycloak customization happens inside its theming and authentication flow system, which is fast for small changes and awkward for radical redesigns. Ory hands you flow data and expects you to build the interface, which is more work upfront but removes ceilings later.
Can I run Ory without operating all four services?
Yes. Many teams start with Kratos for identity and Hydra for tokens, then add Keto or Oathkeeper only when a real need appears. Each service is independently useful, so you can grow the stack as requirements grow rather than deploying everything on day one.
Does Zitadel replace the need for a separate authorization service?
Often, yes. Zitadel’s project-level roles and API-first design cover many application authorization needs without a separate engine. Teams with complex relationship-based rules may still want a dedicated system, but simple role checks usually stay inside Zitadel.
Which platform is best for a SaaS product with many customer tenants?
Zitadel’s organizations are built for exactly this pattern, so it usually requires the least custom glue. Keycloak can do it with a realm per tenant, though that multiplies configuration. Ory leaves tenancy modeling to your application layer, which is flexible but more work.
How risky are upgrades on each platform?
All three require planning. Keycloak’s single-server model makes upgrades visible but occasionally breaking in themes and APIs. Ory’s independent services need a compatibility matrix. Zitadel’s event-sourced core rewards rehearsing schema changes in staging before production.
Should I decide based on 2027 roadmaps?
No. Roadmaps shift, and predictions about passkeys, verifiable credentials, or AI-agent identity leadership are speculative. Decide on the workflow you can operate today, confirm current licensing per component, and revisit the choice annually as the projects evolve.


