Serverless File Storage

Serverless functions are stateless and short-lived, which makes file handling awkward in ways that bite you in production. A Lambda's /tmp is ephemeral and capped, function execution time is limited, and — the classic trap — any client that holds connections (a database driver, an S3 client with a connection pool) fights the concurrency model, because a thousand simultaneous invocations try to open a thousand pools.

The clean pattern for serverless storage is a stateless HTTP call to an external service, and that is exactly the shape files.link takes. There is no connection pool to warm, no SDK whose cold-start initialization adds latency to the first invocation, and no driver to keep alive between requests — just three HTTP calls: POST /v1/files/{folderId} with your raw API key in the Authorization header returns a presigned URL; PUT the bytes; POST /v1/files/confirm-upload to finalize and get the permanent CDN URL.

There is a serverless-specific optimization hiding in that flow. Because step one returns a presigned URL, your function doesn't have to be the thing that streams the file body at all. A common pattern: a tiny function calls POST /v1/files/{folderId}, returns the presigned URL to the browser, and the browser PUTs the bytes directly — so a large upload never passes through your function, never counts against its execution time or memory, and never risks hitting the API Gateway payload limit.

The function only brokers a short JSON request and the lightweight confirm call. This works from anything that speaks HTTP: AWS Lambda, Google Cloud Functions, Azure Functions, Vercel Functions, Netlify Functions, Cloudflare Workers, Deno Deploy. Edge runtimes that forbid Node built-ins and native modules can't run many storage SDKs at all, but fetch() is always available, so the three-call flow runs at the edge unchanged.

Files are encrypted at rest and move over TLS; private files are gated by signed URLs your function mints per request. And because billing is prepaid credits rather than per-request metering, a function that fans out to high concurrency draws down a balance you control instead of producing a request-count invoice you discover at month end.