Engine troubleshooting
Fix missing CLIs, failed authentication, model errors, and path differences.
“CLI not found”
- Open a terminal and run the CLI directly.
- Confirm the command is an executable, not only a shell alias or function.
- Restart MengBot after installation.
- Set the exact executable in Settings → Engines if automatic detection still fails.
Desktop applications often inherit a different environment than a terminal. This is common on Windows shortcuts, macOS Finder launches, and Linux application menus.
Authentication failures
Run the provider's own status or login command in a terminal. MengBot uses the CLI's authenticated session; installing the binary alone is not enough.
A 401 Unauthorized response from api.openai.com means the launched Codex process did not receive valid authentication. Confirm codex login status, sign in again if needed, then restart MengBot. Do not paste a ChatGPT password into MengBot.
A selected model fails
- Refresh the model picker after updating the CLI.
- Choose a model actually available to that provider account.
- Confirm a custom wrapper forwards the model argument and preserves the provider's expected environment.
- Try the provider CLI directly with the same model to separate an account problem from an MengBot integration problem.
Codex sandbox mismatch
MengBot preserves Codex's resolved network and writable-folder settings when starting a turn. If Codex returns no resolved sandbox or a different mode from the one requested, the turn stops before sending your message. Check the native Codex sandbox setup and selected access mode, then retry. MengBot does not silently replace that policy with broader access.
Windows notes
Use a real .exe, .cmd, or resolvable command path in the engine override. If a CLI was installed in WSL, the Windows desktop app cannot treat that Linux binary as a native Windows executable without an explicit bridge or wrapper.