younifyd
← All articles
Engineering11 min read

younifyd's connector architecture: how we built 50+ integrations without an SDK

y

The younifyd team

July 3, 2026 · younifyd.com

Building connectors at scale means solving auth, rate limiting, pagination, error handling, and credential rotation once — not per connector. When we set out to build 50+ integrations, we made a deliberate choice: define a connector contract and let the platform handle the cross-cutting concerns. Here's the architecture decision that let us ship 50 connectors in four months.

The connector contract

Every younifyd connector is a YAML manifest that declares: the service's base URL, authentication scheme (OAuth, API key, basic, signed JWT), rate limit headers to respect, pagination strategy, and a catalogue of operations with their input/output schemas. The platform reads the manifest and handles everything else — token refresh, retry on 429, cursor-based pagination. The connector author never writes retry or pagination logic.

# connector manifest: stripe.yaml
name: stripe
base_url: https://api.stripe.com/v1
auth: { type: bearer, header: Authorization }
rate_limit: { header: X-RateLimit-Remaining }
retry: { on: [429, 503], backoff: exponential }
pagination: { type: cursor, field: starting_after }

Auth handled by the platform

Auth is the hardest part of building integrations — not because OAuth is complicated in isolation, but because you need to handle token expiry, refresh token rotation, credential storage, and per-user credentials at scale. younifyd's connector layer manages all of this. When a user connects a service, their credentials are stored encrypted per-workspace. The platform refreshes tokens automatically and handles credential rotation without any connector-specific code.

Rate limit awareness

Many integration platforms ignore rate limit headers until a connector gets 429'd repeatedly. younifyd reads the Retry-After or X-RateLimit-Remaining header on every response and adjusts the concurrency of calls automatically. For connectors that don't return rate limit headers, we apply conservative defaults and back off exponentially on 429s. This alone eliminated a class of production incidents we saw in our early connectors.

The speed advantage

Because the platform handles auth, pagination, rate limits, retry, and error normalization, adding a new connector is a few hours of work, not weeks. We define the manifest, add the operation catalogue, and write integration tests against the real API. The most complex connectors — Salesforce, HubSpot, Stripe — took two to three days. Simple REST APIs take half a day. That's how we shipped 50 connectors in four months with a team of five engineers.

Use 50+ connectors in your workflows

All connectors are included in every plan. Request a custom connector through Enterprise.

Try for Free