Description
When using large local models via Ollama (e.g., deepseek-coder:33b or devstral-small-2:24b), Wingman consistently fails to complete requests that take longer than 5 minutes. The extension returns a 500 error and cancels the request after exactly 5 minutes, regardless of whether the model is still generating a response. This makes it impossible to use large local models for any non-trivial task.
Steps to Reproduce
Set up Wingman-AI with Ollama as the AI Provider.
Select a large model: deepseek-coder:33b or devstral-small-2:24b.
Send a request that requires significant processing time (e.g., "Write a complete Python script to download a YouTube video and set a custom thumbnail based on a timestamp").
Wait 5 minutes.
Observe the error in the chat: I was unable to continue, reason: fetch failed.
Check the logs — a 500 status code appears exactly at the 5-minute mark, followed by cancel task.
Expected Behavior
Wingman should either:
Respect a user-configurable timeout value.
Wait indefinitely for the model to complete its response.
Provide a clear setting to adjust the request timeout.
Actual Behavior
Requests are forcibly cancelled after exactly 5 minutes.
No user-configurable timeout option is available in the settings UI or settings.json.
The environment variable OLLAMA_LOAD_TIMEOUT has no effect on Wingman's behavior.
The error message fetch failed is misleading and does not clearly indicate a timeout issue.
Environment
- VSCode Version: 1.128.0-insider (user setup)
- Wingman-AI Version: 1.5.3
- Operating System: windows 10
- AI Provider being used: Ollama
Logs
[GIN] 2026/07/04 - 01:33:16 | 500 | 5m0s | 127.0.0.1 | POST "/api/chat"
srv stop: cancel task, id_task = 149
slot release: id 0 | task 149 | stop processing: n_tokens = 2605, truncated = 0
Description
When using large local models via Ollama (e.g., deepseek-coder:33b or devstral-small-2:24b), Wingman consistently fails to complete requests that take longer than 5 minutes. The extension returns a 500 error and cancels the request after exactly 5 minutes, regardless of whether the model is still generating a response. This makes it impossible to use large local models for any non-trivial task.
Steps to Reproduce
Set up Wingman-AI with Ollama as the AI Provider.
Select a large model: deepseek-coder:33b or devstral-small-2:24b.
Send a request that requires significant processing time (e.g., "Write a complete Python script to download a YouTube video and set a custom thumbnail based on a timestamp").
Wait 5 minutes.
Observe the error in the chat: I was unable to continue, reason: fetch failed.
Check the logs — a 500 status code appears exactly at the 5-minute mark, followed by cancel task.
Expected Behavior
Wingman should either:
Respect a user-configurable timeout value.
Wait indefinitely for the model to complete its response.
Provide a clear setting to adjust the request timeout.
Actual Behavior
Requests are forcibly cancelled after exactly 5 minutes.
No user-configurable timeout option is available in the settings UI or settings.json.
The environment variable OLLAMA_LOAD_TIMEOUT has no effect on Wingman's behavior.
The error message fetch failed is misleading and does not clearly indicate a timeout issue.
Environment
Logs
[GIN] 2026/07/04 - 01:33:16 | 500 | 5m0s | 127.0.0.1 | POST "/api/chat"
srv stop: cancel task, id_task = 149
slot release: id 0 | task 149 | stop processing: n_tokens = 2605, truncated = 0