Updated November 10, 2026
DeepSeek SuperAgent on Windows: WSL vs Native Windows Setup (November 2026)
Choose a Windows route, understand the limits, and check your environment before installing.
For a Windows machine, WSL 2 gives the prototype a Linux environment closer to the published Unix-oriented setup. Native Windows has community-reported compatibility issues; third-party ports are separate projects. WSL helps with compatibility, but it is not a security boundary for untrusted agent work.
DeepSeek SuperAgent is a research prototype built on DeepSeek Harness, not an official DeepSeek product or production-ready software. It needs a separately configured model provider. This guide follows the published documentation; we have not run an end-to-end deployment for this article.
DeepSeek Harness running on Windows does not by itself establish that every SuperAgent service script and worker flow runs there. Choose the environment for the whole workflow: installation, GitHub credentials, per-ticket worktrees, service supervision and review.
Should you choose WSL 2, native Windows, or a separate Linux host?
| Route | What it offers | What to check |
|---|---|---|
| WSL 2 | Linux tools and filesystem on a Windows machine; closer to the published tmux and Python workflow. | Windows mounts, interop, credentials and networking can still expose host resources. We have not tested this deployment end to end. |
| Native Windows | Windows-native tools and desktop workflow. | Community reports describe service-script, path and worker-tool issues in particular versions. Do not assume a patched setup applies to your version. |
| Separate Linux host or VM | A more distinct place to manage the runtime and credentials. | Isolation depends on VM configuration, shared folders, network access and credentials. A VM is not automatically safe. |
Why is the original setup awkward on native Windows?
The published service workflow uses Python and tmux. A Windows user reported that the service script imports the Unix-only fcntl module, that some directory validation rejected Windows drive-letter paths, and that worker sessions lacked expected tools in their particular setup. That report used Windows 11 and DSH CLI 0.2.1-rc.3; it does not prove every version has the same failures or that its local patches are maintained fixes.
Community Windows ports and wrappers can change those assumptions. Treat them as third-party implementations, review their code and permissions, and check their version and maintenance status before running them. We have not independently tested a native port for this guide.
How do you prepare WSL 2?
On a supported Windows installation, open PowerShell as an administrator and use Microsoft’s installation command:
wsl --installComplete any requested restart and finish creating the Linux user. From PowerShell, wsl --list --verbose shows installed distributions and their WSL version. If WSL is already installed, inspect its current state first rather than assuming the command creates a fresh isolated environment.
Inside the Linux distribution, use Linux versions of Node.js, pnpm, Git, Python, tmux, the GitHub CLI and DeepSeek Harness. Keep the working clone on the Linux filesystem when running Linux tools; cross-filesystem access can be slower. Do not place personal tokens or production secrets in the service user’s home directory.
How do you keep the Windows host and test data apart?
WSL can interact with Windows and may access mounted Windows drives. A separate Linux username or a repository under the Linux home directory reduces accidental mixing but does not, on its own, stop access to the host. For untrusted code, use a separately managed environment with deliberate isolation, limited networking and credentials that are safe to revoke. Check any shared folders and mounts before granting a worker access.
Inside that environment, create a dedicated GitHub user for the agent and a separate CLI configuration. Start with a test repository, one allowed requester and branch protection. The worker executes with the service user’s OS permissions, so access should be limited at the OS and infrastructure level as well.
What is the WSL setup sequence?
- Open the Linux shell and check
node -v,pnpm -v,git --version,python3 --version,tmux -V,gh --versionanddsh --version. - Check the prototype’s current prerequisites and the Harness version range. Configure a model provider and a dedicated agent GitHub CLI account within the environment.
- Clone and build the prototype, create the dedicated profile for a new installation, and install the bundle. Do not run the profile-creation command against an existing profile because it replaces its configuration.
- Configure the local panel, GitHub Source, allowed requester and executor’s worktree preparation. Start in
observefor inferred candidates, with human review and automatic merging disabled. - Check
scripts/dsh-superagent status, then use a small explicit request in the test repository. Explicit requests may execute in observe mode.
For the full issue-to-review flow, use the existing SuperAgent setup and cost guide. A successful dependency check is only a starting point; verify ticket creation, the workspace, permissions and the resulting diff separately.
What should you check when something fails?
| Symptom | What to inspect |
|---|---|
tmux or Python error | Confirm you are running the Linux service script inside WSL, with the required Linux tools installed. |
| No GitHub response | Check the agent wrapper’s login, repository access, allowed requester and source status. |
| Ticket cannot prepare checkout | Check that the Linux clone path exists and the prepare command uses the correct filesystem and credentials. |
| Worker lacks expected tools | Record DSH and bundle versions and inspect the worker preset. A Windows report observed this in one version; it is not a universal diagnosis. |
| Panel cannot reach the service | Check whether the process is running and where the panel is bound. Avoid exposing a local management panel to a network without appropriate controls. |
Frequently asked questions
Is WSL a sandbox?
No. WSL and Windows can interoperate; configure real isolation separately for untrusted work.
Does this guide prove native Windows support?
No. It describes published requirements and individual community reports, not an end-to-end native Windows test.
Can I use my everyday GitHub account as the agent?
The project recommends a dedicated agent account. Separate it from your normal identity and grant only access needed for the test.