goose randomly stops responding

[albina-popova]

Describe the bug

goose becomes sometimes unresponsive. I can send new messages in the same chat windown, but I am not getting any answers back.

Changing the model does not appear to fix the situation, only starting the new chat helps.

diagnostics_20251122_4.zip

diagnostics_20251122_2.zip

---

To Reproduce
Steps to reproduce the behavior:

1. Go to goose macos desktop app
2. start a new chat using Github Copilot -> gpt-5-mini
3. notice that goose stops responding

---

Expected behavior
goose should continue replying whenever prompted

---

Screenshots

---

Please provide the following information

• OS & Arch: macOS 15.6
• Interface: UI
• Version: 1.15.0
• Extensions enabled: Chatrecall, extension manager, developer, todo, brave web search
• Provider & Model: Github Copilot - gpt-5-mini

---

Additional context
Diagnostics file attached.

[michaelneale]

thanks @albina-popova from the diagnostics - the LLM from github copilot is returning nothing back, nothing goose can do with that, so that is interesting. Do you have any non github copilot providers to try with for the same model (although I assume you tried non mini - and saw the same).

Some providers will do things like this when there is load shedding, or some hidden rate limiting (especially ones that aren't pay per token) so that is likely what is happening here (but not in a way that it can back off and retry). But then again, a new session makes be wonder if some other limit is being hit which again, nothing shows up with.

There have been some recent fixes around these models and also github copilot with session/refreshing - so I suspect that may be related. Can you confirm if this happened with sonnet - not just the gpt family?

[albina-popova]

Thank you for the quick reply. I can confirm that the issue reappears if I were to change the model to one of github copilot non-gpt models, e.g. claude haiku. Yet, if I were to switch to openrouter within the same chat, the replies from goose would resume. However, I have a feeling that this is not connected with rate limiting. I have been trying to reproduce the same issue with earlier version of goose or another general open source agent Cherry studio with the same llm provider-model combination and was not able to.

[michaelneale]

@albina-popova when you say a new chat - do you mean a new window?

also - can you tell me what version of goose it works with? (ideally and then when stopped) so we can bisect. There were some recent fixes so that likely will help too.

[albina-popova]

@michaelneale apologies for the delay with a response.

With goose Version 1.12.0 when trying to resume the same chat that has previously become unresponsive in 1.15.0 version, I get an error response below, which might be the root cause of this bug.

"Ran into this error: Request failed: image urls not allowed in non-user messages (code: invalid_request_body) (status 400).

Please retry if you think this is a transient or recoverable error."

diagnostics_20251129_3.zip

If not trying to resume the chat with 1.12.0 that became unresponsive with 1.15.0, but using 1.12.0 for any new prompts not referring to images and relying on Github CoPilot -> gpt-mini-5 I do not encounter the same issue with unresponsiveness.

I have been testing recently the newly added ability of goose to process images with the Github Copilot as a LLM provider and this may have been a constant in all of the cases where I have encountered an issue with lack of responsiveless, although I cannot say with 100% confidence.

[nickytonline]

I ran into this a couple of days ago with GitHub Copilot models in Goose @albina-popova. I believe the scenario is you've used up your GitHub Copilot premium requests, or at least that's what happened for me.

In goose with GitHub Copilot, Claude Opus 4.5 it stops responding. If I use the same premium model in VS Code, I get the premium requests error message. In VS Code it automatically downgrades the model to GPT-4.1 to be able to answer you, but I don't think that happens in Goose because it wouldn't have that Copilot specific mechanism.

I guess to @michaelneale's point, maybe we could chat with GitHub about this.

The LLM from github copilot is returning nothing back, nothing goose can do with that, so that is interesting.

I can ask in the GitHub Stars channel.

[albina-popova]

I am quite sure that this specific issue is not connected with rate limiting. As i can start a new chat right away with the same provider/model combination and that would work. Also I have not experienced this issue with Goose v1.12 at all.

[yohanpainchaud-blip]

Hello,
I see also the same behavior with version 1.30.0 on Windows11 installation, and with gemini_pro_3 LLM.
After the following message in my session, Goose doesn't show me any answer:

BUT I can see in the log files that it produces an answer, but don't display it to me ??!?