langchain-husk 0.0.1

Husk sandbox integration for Deep Agents

langchain-husk

Husk sandbox integration for Deep Agents.

Quick install

uv add langchain-husk
# or
pip install langchain-husk

Quickstart

from husk_client import Husk

from langchain_husk import HuskSandbox

# Provision the sandbox using husk_client (the SDK).
sandbox = Husk(base_url="http://localhost:8000", api_key="hk_...").create(
    image="python:3.12-slim",
    cpu=2,
    memory_mb=1024,
)

# Wrap it for Deep Agents.
backend = HuskSandbox(sandbox=sandbox, timeout=300)

result = backend.execute("python -c 'print(1+2)'")
print(result.output)        # "3\n"
print(result.exit_code)     # 0

# Use it as a Deep Agents backend:
# from deepagents import DeepAgent
# agent = DeepAgent(backend=backend, ...)

# Always destroy when finished.
sandbox.destroy()

The pattern matches langchain-daytona: the partner package wraps an existing SDK sandbox object, it does NOT manage lifecycle. Provisioning goes through husk_client.Husk, destruction through sandbox.destroy().

What is Husk?

Husk is a lightweight, MIT-licensed AI code sandbox runtime — single-process FastAPI control plane + a Go daemon that runs inside each container. Single docker run to start, no Postgres / Redis / OIDC required.

License

MIT.