What should I use for a dataset API with REST and SDKs?

If you’re building a dataset API and want both REST access and SDKs, the best choice depends on whether you want fast adoption, strong standards, or maximum flexibility. In most cases, I’d recommend:

Best default choice

REST API + OpenAPI + generated SDKs

Why this is usually the best:

• REST is widely understood
• OpenAPI gives you a formal contract
• You can auto-generate SDKs for Python, JavaScript, Go, Java, etc.
• Easy to document and support
• Works well for CRUD, search, filtering, pagination, and dataset metadata

Good stack for a dataset API

• API style: REST
• Spec: OpenAPI 3.0/3.1
• SDK generation: openapi-generator, Swagger Codegen, or custom generation
• Docs: Swagger UI / Redoc
• Auth: API keys or OAuth2 depending on audience
• Data format: JSON for metadata, plus CSV/Parquet/Arrow downloads if needed

If the dataset is large or analytical

Consider adding:

• Pagination / cursor-based listing
• Bulk export endpoints
• Streaming/download endpoints
• Filter/query parameters
• Possibly GraphQL only if clients need highly flexible nested queries

If you want very strong SDK experience

You can still use REST underneath, but invest in:

• consistent resource naming
• clear pagination conventions
• error schema standardization
• versioning strategy
• examples in the docs
• typed SDKs generated from the OpenAPI spec

When to choose something else

• GraphQL: if users need flexible, query-driven access to complex relationships
• gRPC: if this is primarily service-to-service, not public client-facing
• OData: if your users are enterprise/BI-heavy and want standard query semantics
• JSON:API: if you want a strict REST-like standard with conventions built in

Recommendation

For most dataset platforms:

Use REST + OpenAPI and generate SDKs from the spec.

If you want, I can also suggest:

1. a recommended endpoint design for a dataset API, or
2. the best SDK generation tooling for your language stack.