[BUG] Roo times out with LM Studio after ~5 minutes (cannot disable; long tasks cut off)

[mattlxdesign]

Type

Bug

Problem / Value

Roo ends long LM Studio requests around 5 minutes even when a higher timeout is set or when timeout is set to 0. This interrupts long coding tasks and wastes time with retries.

Context

Affects users running long tasks with LM Studio or other local/OpenAI-compatible providers (e.g., Orchestrator mode, large 128k context). Setting apiRequestTimeout above 300s does not extend beyond ~5 minutes; setting 0 leads to immediate timeout. Provider-specific timeout fields do not persist after export/import.

Reproduction

1 Use Roo Code v3.25.23–v3.26.1 with LM Studio and run a task >5 minutes (e.g., Orchestrator switching modes).
2) Set apiRequestTimeout to >600s or to 0.
3) Run the task and wait past 5 minutes.
Expected: Request runs until completion; 0 disables timeout; provider timeouts persist and apply.
Actual: Request times out at ~5 minutes; 0 causes immediate timeout; provider timeout key is not retained.

Constraints/Preferences

• Option to disable timeouts for local models.
• Ensure per-provider timeouts persist and match docs.
• Avoid stacking retries while a prior request is still running.
  EOF
  )

[daniel-lxs]

Hey @mattlxdesign Thank you for reporting this issue, I think your solution of disabling timeouts when setting it to 0 makes sense, however it might be a bit hard for me to test if the fix works.