dsh-tensorlake-sandbox DSH plugin

@tensorlakeai/dsh-sandbox is an installable DSH bundle that replaces the host subprocess and filesystem providers with remote Tensorlake sandbox counterparts. It runs in a microVM with configurable CPU, memory, disk, and timeout, and does not require changes to the Harness installation. The bundle keeps the bash-sandbox provider in danger-full-access mode while delegating execution to the Tensorlake runtime.

remote-execution developer-tools

How to install the dsh-tensorlake-sandbox DSH plugin

dsh plugin --profile headless add @tensorlakeai/dsh-sandbox

Copying does not run this command. Review the repository and version before installing the dsh-tensorlake-sandbox DSH plugin.

dsh-tensorlake-sandbox DSH plugin data source

dsh-tensorlake-sandbox DSH plugin snapshot date: Aug 16, 2026

discovered

What the dsh-tensorlake-sandbox DSH plugin can do

Replaces host subprocess and fs-sandbox providers with Tensorlake remote sandbox providers
Supports configurable CPU, memory, disk, and inactivity timeout via Cordis config
Verifies that the configured working directory (cwd) is writable and uses sudo to create a protected path when needed
Keeps bash-sandbox mounted in danger-full-access mode, satisfying the permission-preset capability contract
Derives both runtime cwd and policy workspace from a single environment variable DSH_TENSORLAKE_CWD
Does not copy credential-shaped environment variables into sandbox processes

Where the dsh-tensorlake-sandbox DSH plugin fits

Build and test code in an isolated, short-lived Linux environment
Run commands that require a consistent Linux filesystem and toolchain without local setup
Execute untrusted scripts or commands in a sandboxed microVM with resource limits
Perform remote development tasks where the host machine lacks the necessary runtime (e.g., Node.js, bash)

Who the dsh-tensorlake-sandbox DSH plugin is for

Developers using DeepSeek Harness who need a secure, remote sandbox for file and process operations
Teams that want to offload task execution to Tensorlake infrastructure while keeping Harness as the orchestrator

dsh-tensorlake-sandbox DSH plugin limitations

Depends on Tensorlake SDK 0.5.103 which pins transitive dependencies with high-severity advisories (undici, nanoid); review upstream advisories before production use
Requires a Tensorlake project with TENSORLAKE_API_KEY and a DEEPSEEK_API_KEY set in the host environment
Only supports the managed Ubuntu image that provides bash, Node.js, and specific GNU tools; custom images must include the same prerequisites
The bundle is designed for @deepseek-ai/dsh 0.1.0-rc.6 or later; older versions are not compatible

dsh-tensorlake-sandbox DSH plugin: from the repository README

Quoted from the tensorlakeai/dsh-tensorlake-sandbox README, the upstream source of the dsh-tensorlake-sandbox DSH plugin. Copyright remains with the original authors.

`@tensorlakeai/dsh-sandbox` moves DeepSeek Harness file, subprocess, Bash, terminal, and LSP operations into one short-lived Tensorlake microVM. It is an installable dsh bundle and does not require changes to the Harness installation. ## Prerequisites - Node.js `^22.19.0` or `>=24.0.0` - `@deepseek-ai/dsh` `0.1.0-rc.6` or a later compatible release - A Tensorlake project with `TENSORLAKE_API_KEY` set in the host environment - `DEEPSEEK_API_KEY` set in the host environment for the default DeepSeek model provider Keep credentials in environment variables or a secret manager; do not commit them to the profile or repository. ## Install Install dsh and add this bundle to the profile you run: ```sh npm install --global @deepseek-ai/dsh dsh plugin --profile headless add @tensorlakeai/dsh-sandbox TENSORLAKE_API_KEY=... DEEPSEEK_API_KEY=... dsh --profile headless "build and test this repo" ``` During development, install a local checkout from its directory: ```sh npm install npm run build dsh plugin --profile headless add . ``` Use `dsh --profile headless --dump-config` to verify that the `@tensorlakeai/dsh-sandbox` layer disables the host `subprocess` and `fs-sandbox` providers, i

Read the full READMERepository license: MIT

dsh-tensorlake-sandbox DSH plugin questions

How do I install the @tensorlakeai/dsh-sandbox plugin?

First install DeepSeek Harness globally: `npm install --global @deepseek-ai/dsh`. Then add the bundle to your profile: `dsh plugin --profile headless add @tensorlakeai/dsh-sandbox`. Make sure you have TENSORLAKE_API_KEY and DEEPSEEK_API_KEY set in the environment.

What does this plugin do?

It replaces the local subprocess and filesystem providers in DeepSeek Harness with remote Tensorlake microVM sandboxes. This allows file operations, Bash commands, terminal sessions, and LSP interactions to run inside a short-lived, isolated Linux environment instead of on the host machine.

What environment variables are required?

You need TENSORLAKE_API_KEY for the Tensorlake project and DEEPSEEK_API_KEY for the default DeepSeek model provider. Keep these in environment variables or a secret manager; do not commit them to the profile or repository.

How can I configure the working directory?

Set the DSH_TENSORLAKE_CWD environment variable to the desired absolute Linux path. The bundle will use this as both the runtime cwd and the policy workspace. You can also override via Cordis config in the profile's cordis.patch.yml file.

What are the known limitations?

The Tensorlake SDK 0.5.103 pins transitive dependencies with high-severity advisories (undici, nanoid). Review upstream advisories before production use. The plugin also requires a Tensorlake project, specific Node.js versions, and a compatible dsh release (0.1.0-rc.6 or later).