Almost all our apps were private. Cloudflare Access already gave us everything we needed, including custom access pages and passkeys. For public apps, we already had Clerk and WorkOS. We didn't need to keep maintaining our own identity platform.
Before Self-hosted Better AuthAfter Cloudflare Access
The reasons
07
01
Cloudflare Access already did everything we needed.
Almost all of our applications were private. Cloudflare Access was the right tool, especially after it added customizable access pages and passkey support. We no longer needed to build our own login experience.
02
Clerk and WorkOS covered public apps.
We already had dedicated authentication options for public-facing products. Running a separate self-hosted identity platform made even less sense.
03
The UI was a nightmare.
We had to build and maintain sign-in, account management, recovery, passkey setup, and admin screens. Auth was supposed to support our apps, not become another full product.
04
We were maintaining our own SCIM.
User and group provisioning, organization-scoped permissions, mappings, security checks, and tests were custom work. The SCIM implementation file alone was 365 lines.
05
API keys and RBAC became a mess.
We built key management, scoped permissions, roles, organization boundaries, revocation rules, and admin controls. More security-critical code we had to maintain ourselves.
06
Cross-domain SSO was extra complexity.
Shared cookies under inds.space did not work across other root domains such as hackthe.world. Those apps needed OAuth/OIDC, redirects, token handling, and separate sessions.
07
We were carrying fork, patch, and security maintenance.
Beyond our application code, we also had upstream customization and compatibility work. Every auth change meant more testing, upgrades, and chances to break access across apps.
How much custom code?
Archived source audit
8,395Physical lines
7,679Nonblank lines
65Source / test / config files
Measured from the final archived source snapshot. Includes our backend, frontends, gateway, migrations, tests, and configuration. Excludes Better Auth itself, dependencies, and generated bundles. The exact upstream fork delta was not independently measurable, so it isn't claimed in these totals.
We didn't abandon authentication. We stopped building what Cloudflare Access, Clerk, and WorkOS already solved for us.
This is a retrospective on our implementation and its maintenance cost, not a claim that Better Auth itself was defective.