LangSmith sandboxes are now a first-class environment in @harborframework 🚀
Install: pip install “harbor[langsmith]”
Authenticate: export LANGSMITH_API_KEY= “<LANGSMITH_API_KEY>”
That's it. Your evals are now ready to run in LangSmith production sandboxes, alongside Daytona, E2B, and Modal.
Users praise LangSmith sandboxes as first-class in Harbor Framework for seamless setup that just works and enables reproducible multi-runtime evaluations without vendor lock-in.
No Digg Deeper questions have been answered for this story yet.
A few things that went into this:
→ Full Dockerfile snapshot support: point Harbor at your task's environment/Dockerfile and it builds, caches, and reuses snapshots automatically → SDK profile support: you can target different environments without config changes
→ Entire lifecycle: exec/upload/download lifecycle fully implemented, so Harbor tasks work the same way regardless of which sandbox provider you're on
If you're running agent evals and want to try it, here are the docs to get started:
https://docs.langchain.com/langsmith/sandbox-harbor
More integrations with LangChain coming soon!
LangSmith sandboxes are now a first-class environment in @harborframework 🚀
Install: pip install “harbor[langsmith]”
Authenticate: export LANGSMITH_API_KEY= “<LANGSMITH_API_KEY>”
That's it. Your evals are now ready to run in LangSmith production sandboxes, alongside Daytona, E2B, and Modal.
@LangChain @harborframework The real value here goes beyond integration—consistent sandbox isolation enables true eval reproducibility across runtimes. Multi-runtime support prevents lock-in while maintaining benchmark integrity. That's what moves the ecosystem forward.
@LangChain @harborframework This is how it should be honestly. Install it, auth, and it just works
@nana_tourSVT @LangChain @harborframework The multi-runtime approach is smart. Prevents vendor lock-in while keeping benchmarks fair. That's a rare combination in dev tools.
@nana_tourSVT @LangChain @harborframework This is exactly what the evals space needed—reproducible benchmarks across runtimes. How do you see this impacting LLM evaluation standards in the next year?
