OpenAI's Windows app documentation, checked October 24, 2026, describes a Windows-native agent by default and a separate WSL option. The agent environment takes effect after restarting the app. The terminal's default shell affects new terminal sessions.
If a command works in your terminal but the agent cannot find the same tool, compare those environments before reinstalling it. The same app window can contain processes using different runtimes.
Identify the two execution environments
Consider this authored setup: a project is in C:\work\analysis, and Python is installed inside an Ubuntu WSL2 distribution. The integrated terminal uses that distribution and can run python3. The app's agent setting is still Windows native.
The successful terminal command establishes that Python exists in that Linux environment. It does not establish that a Windows PowerShell process has a native python command. Asking ChatGPT to retry the same command without that distinction can repeat the same failure.
Record these four fields before requesting a repair:
| Field | Fictional setup |
|---|---|
| Project location in Windows | C:\work\analysis |
| Same Windows location viewed from WSL | /mnt/c/work/analysis |
| Agent environment | Windows native |
| Integrated terminal | Ubuntu WSL2 |
These are illustrative paths and settings, not observations from your computer. Do not move the project just to match the example. The first question is which runtime the task actually needs.
Confirm the environment without installing anything
In a PowerShell session, these read-only commands show the shell, current folder and available Python command matches:
$PSVersionTable.PSEdition
Get-Location
Get-Command python, python3 -ErrorAction SilentlyContinue
Inside the intended WSL terminal, inspect the corresponding context:
uname -s
pwd
command -v python3
Run each group in the named environment; pasting the Linux commands into an unrelated Windows shell changes the test. Preserve the actual output, including an absent command. Do not describe one environment's result as if it came from the other.
Ask the agent to report its own execution environment with a similarly bounded inspection. The terminal's output alone cannot settle where the agent executed its failing command.
Change the setting that matches the intended workflow
If you intend to run the agent in WSL2, use the app's agent environment setting and restart the app. Then check the agent's environment again before retrying the project command. If you intend to keep a native Windows agent, inspect the native tool requirements and paths for that workflow instead.
OpenAI documents that you can keep a WSL agent and PowerShell terminal, or use WSL for both. It also recommends Windows-filesystem project storage when keeping the Windows-native agent, with WSL access through /mnt/<drive>/.... Opening a WSL filesystem project and selecting a WSL agent are separate choices. Windows and WSL configuration
An environment change is not a reason to install another package manager, replace dependencies or weaken permissions. Establish whether the current tools already exist in the intended runtime first. Also check which WSL distribution contains them if more than one is installed.
Send a diagnosis request with both settings
Help diagnose a command mismatch in the ChatGPT Windows desktop app.
App version: [ENTER]
Agent environment setting: [WINDOWS NATIVE / WSL]
Integrated terminal setting: [ENTER]
WSL distribution/version, if relevant: [ENTER]
Project location as seen from each environment: [ENTER]
Exact failing command, where it ran and full output: [PASTE]
Same-command result in the integrated terminal: [PASTE]
Read-only shell/path/tool checks already performed: [PASTE]
First explain which environment each observation describes. Identify
the smallest additional read-only check that distinguishes a missing
tool from a wrong shell, distribution or path. Do not install software,
move files, change permissions or infer that terminal success proves
agent access. Explain any proposed setting change and when it takes
effect before asking me to retry the original command.
For the fictional Python setup, the required answer is a consistent choice of runtime and project path. It is not necessarily a second Python installation. The Python guide covers checking a script once the interpreter is established, and the shell-script guide covers a Bash script's argument rules.
Retry the original command in the confirmed agent environment and record its actual result. Keep that environment choice with the project's setup notes so a terminal preference change cannot be mistaken for a runtime migration later.