> For the complete documentation index, see [llms.txt](https://docs.zenml.io/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.zenml.io/stacks/stack-components/sandboxes/local.md).

# Local

A no-isolation subprocess sandbox for development and examples.

The local sandbox flavor runs commands as `subprocess.Popen` on the same machine as the step, with the same OS user, full filesystem access, and full network access. **It provides no isolation.**

It is intended for:

* Quickstart / examples that should work without a Modal or other cloud account.
* Unit tests of code that consumes the `BaseSandbox` interface.
* Local development against the abstraction.

It is **not** intended for running untrusted LLM-generated code. For real isolation, use a different sandbox flavor.

## How to register

```bash
zenml sandbox register local-sb --flavor=local
zenml stack update --sandbox local-sb
```

## Settings

The Local sandbox inherits `BaseSandboxSettings` and adds `forward_env`.

* `sandbox_environment`: explicit environment variables to set inside the subprocess.
* `forward_env`: which parent-process variables are forwarded into the session. Set `forward_env=True` to forward the entire parent environment, `False` to forward nothing, or pass a list of names to forward exactly those. It defaults to a set that keeps command resolution and UTF-8 stdio working: `PATH`, `HOME`, `LANG`, `LC_ALL`, `TMPDIR`, `TERM`.

Secrets in the parent environment stay isolated from the session unless `forward_env` is `True` or names them explicitly. `sandbox_environment` is layered on top and overrides forwarded variables on a key collision.

## What it doesn't support

| Feature                         | Local | Notes                                                                                                                                                  |
| ------------------------------- | ----- | ------------------------------------------------------------------------------------------------------------------------------------------------------ |
| `snapshot()` / `restore()`      | ❌     | No state-capture primitive locally.                                                                                                                    |
| `attach(session_id)`            | ❌     | Sessions are bound to the parent process; not re-attachable.                                                                                           |
| `upload_file` / `download_file` | ❌     | Files live in the session workdir already; use plain Python file I/O.                                                                                  |
| Streaming output                | ✅     | `subprocess.PIPE` line-buffered.                                                                                                                       |
| Sandbox log forwarding          | ✅     | Each `exec` emits a `$ <command>` marker plus stdout/stderr lines + an `OK`/`FAIL exit code <code>` trailer into the step's `sandbox:<id>` log source. |
| `destroy()`                     | ✅     | Delegates to `close()`; removes the session workdir.                                                                                                   |

## When to use it

Use the local flavor when:

* You want to try the Sandbox abstraction without setting up infrastructure.
* You're running an example that demonstrates the agent / sandbox split but doesn't care about isolation.
* You're writing a unit test that exercises a tool that consumes a `SandboxSession`.

Otherwise, use a containerized flavor.


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://docs.zenml.io/stacks/stack-components/sandboxes/local.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.