Add langchain-sprites partner package (Fly.io Sprites sandbox backend)

[kylemclaren]

Submission checklist

• This is a feature request, not a bug report.
• I searched existing issues and didn't find this feature.
• I checked the docs and README for existing functionality.
• This request applies to this repo (deepagents) and not an external package.

Area (Required)

• deepagents (SDK)
• dcode
• talon
• acp
• evals
• harbor
• daytona
• modal
• quickjs
• runloop
• vercel
• langsmith-sandbox
• Other / not sure / general

Feature description

Feature

Add a langchain-sprites partner package under libs/partners/sprites. The package is a sandbox backend for Fly.io Sprites. It implements the SandboxBackendProtocol through the sprites-py SDK.

Motivation

Sprites are persistent Linux VMs with names. A Sprite starts in milliseconds. A Sprite stops automatically when it is idle, and a stopped Sprite has no cost. A Sprite can make a checkpoint of the full machine state in less than one second and can restore it. These properties are a good fit for Deep Agents: an agent can keep its environment between runs, and a user can roll back the changes an agent makes.

The JavaScript version of this provider is already open: langchain-ai/deepagentsjs#807.

Proposed solution (optional)

Proposal

• SpritesSandbox extends BaseSandbox: execute() with bash -lc, upload_files() and download_files() through the Sprites filesystem API.
• A .sprite property for direct SDK access (checkpoint, restore, reconnect by name).
• Unit tests with a mock SDK, and the SandboxIntegrationTests standard suite as integration tests. The suite passes 79 of 86 tests against the live Sprites API; the 7 failures also occur for the first-party LocalShellBackend (released langchain-tests expects older glob()/write() semantics).
• Repository wiring per libs/partners/AGENTS.md.

Additional context (optional)

A PR is ready: #6001

[mdrxy]

Thanks for putting this together — this is solid work and we appreciate you taking the time to contribute it.

We've decided to stop accepting new sandboxes into the monorepo at this time. Each example we merge becomes ours to maintain: keeping it building against new SDK versions, patching dependencies, responding to issues filed against it, and reviewing its security surface (especially examples that wrap shell execution, browser automation, or third-party APIs with credentials). That overhead has grown faster than our capacity to absorb it, and rather than letting examples bit-rot in-tree we'd rather not take them in the first place.

That said, we'd really like to see this continue as a standalone package — and there are real advantages to that route: you keep full ownership, ship on your own schedule, and aren't coupled to our release process. Here's the path:

1. Publish it to PyPI so users can install it directly.
2. Once published, open a PR to the docs repo (https://github.com/langchain-ai/docs) to get it listed alongside other sandbox integrations — the contributing guide has the requirements and template.

Happy to answer questions if anything in that process is unclear. And we're still working out the best way to recommend community packages to users — feedback welcome there too.