Files SDK 1.3 adds 12 storage adapters and exists() checks

• haydenbleasel's launch post says Files SDK 1.3 shipped with 12 new storage adapters, plus a new exists() helper and a Files.file(key) handler.
• The official website and GitHub repo, both linked from the install post and the source link, frame Files SDK as one file abstraction over multiple blob and object stores.
• LLMpsycho's repo summary adds that the SDK already wrapped S3, R2, GCS, Azure, and Vercel Blob, and exposed prebuilt tools for Vercel AI SDK, OpenAI Agents, and Claude Agent SDK.
• In dotta's storage thread, the launch lands inside a broader agent pattern: teams want local-file semantics even when their jobs run inside stateless sandboxes.

You can browse the refreshed docs site, inspect the open source repo, and compare the release to dotta's survey of agent "disk-like" systems. The interesting bit is not the version bump itself. It is that file APIs are turning into agent infrastructure, with one package trying to sit between object storage, blob storage, and agent toolchains.

Files SDK 1.3 is a small release with three concrete changes: 12 new adapters, exists(), and Files.file(key). The launch post does not enumerate the 12 adapters inline, but the linked website and repository position the package as a storage abstraction layer rather than a single-backend client.

The install surface stayed simple: npm i files-sdk, per haydenbleasel's install post. That matters mostly because adapter-heavy SDKs usually get harder to adopt as they broaden, and this one is still presenting itself as a single-package drop-in.

According to LLMpsycho's repo summary, Files SDK already ships ready-made tools for Vercel AI SDK, OpenAI Agents, and Claude Agent SDK. That turns it from a plain storage wrapper into harness glue for agent stacks that need to read, write, and swap backends without rewriting app code.

The same post says the abstraction covered S3, R2, GCS, Azure, Vercel Blob, and more before 1.3. The new adapter count suggests the release is widening that compatibility layer faster than it is adding headline API surface.

dotta's thread is useful context because it breaks the problem into concrete storage patterns agents keep bumping into:

• cloud-native shared filesystems, such as S3-backed layers
• remote filesystem adapters over object stores
• git-backed workspaces, where each run or attempt can branch and diff
• versioned filesystems for long-lived artifacts
• durable disks bundled with execution environments

That framing makes Files SDK 1.3 look less like another blob wrapper and more like a bet on a common interface for these systems. The thread's core observation, in dotta's writeup, is that agents are trained around local text files, but production jobs often run in stateless sandboxes where that assumption breaks.

The project is open source on GitHub, installable from npm as files-sdk, and had 623 GitHub stars in LLMpsycho's May 13 post. That gives the release a clearer shape: a lightweight package, public source, and a growing adapter matrix aimed at developers building agent workflows on top of commodity storage.