[Scenario]: Terminal Chat

[zadjii-msft]

We'll be using this thread to track the ongoing work for Copilot for the Windows Terminal.

This is not something that's in public preview yet.

More details will follow soon!

Initial PR: #16285

Tasks

Beta Give feedback

1. [Terminal Chat AI] Allow company wide disabling this feature #16401
   Area-Chat Issue-Task Needs-Tag-Fix Product-Terminal +4
2. [Terminal Chat AI] Add run button to execute suggested commands inline #16435
   Area-Chat Issue-Feature Needs-Tag-Fix Product-Terminal +4
3. Attempt to configure "Terminal Chat" on second machine results in invalid Azure OpenAI endpoint error. #16442
   Area-Chat Issue-Bug Needs-Tag-Fix Product-Terminal +4
   
4. Clicking a command in Terminal Chat that contains sudo produces a sudo error #16485
   Area-Chat Issue-Bug Needs-Tag-Fix Priority-3 Product-Terminal +5
5. Windows Terminal Chat does not consistently detect commands in responses from Azure OpenAI #16484
   Area-Chat Issue-Bug Needs-Tag-Fix Product-Terminal +4
   
6. [Terminal Chat] The LLM receives confusing information about shell #18142
   Area-Chat Issue-Bug Needs-Tag-Fix Product-Terminal +4
   

[WeihanLi]

Seemed not working for me, even the network request success

network request && response:

[PankajBhojwani]

Seemed not working for me, even the network request success

Thanks for letting us know about this! What is the endpoint you are using?

[WeihanLi]

The endpoint I use is

https://spark-01.openai.azure.com/openai/deployments/gpt-35-turbo/chat/completions?api-version=2023-05-15

[PankajBhojwani]

@WeihanLi do you have any content filters on for that endpoint? We expect the endpoint to have content filters on with severity = safe

[WeihanLi]

do you have any content filters on for that endpoint?

I think so, it may relate to this. I'm using the default settings, the content filter could exists

[PankajBhojwani]

Could you check that? I think your content filters might be switched off, just based on the screenshot you sent. With the content filters on, the JSON response should contain a prompt_filter_results or prompt_annotations field

[WeihanLi]

No item for the content filter page

Are there more places I need to check?

[mikenelson-io]

Same error here now on any deployments. Even my old deployments no longer work and get this error.

There does not appear to be any way to set any of the content filtering to "safe" via the Playground UI. You can only set Low-Medium-High.

[g0t4]

@mikenelson-io I had that too so I went into content filters and turned everything on and highest level and then I cleared credentials in win term and it worked...

[g0t4]

The troubleshooting experience could really use some ❤️... i.e. logging more info on failures

[ChrisYin]

@PankajBhojwani I see you PR for supporting OpenAI in Terminal Chat: #17540
Could you please be considering support self-hosting LLM models?
Since different model may use different request headers.
A similar feature has been supported in an extension of VScode, see: https://docs.continue.dev/setup/configuration, Section: Self-hosting an open-source model
User can specify its own request headers by setting config.json.
Looking for Terminal Chat can support more self-hosting opensource models.

[PankajBhojwani]

Could you please be considering support self-hosting LLM models?

@nguyen-dows thoughts on this?

[mikenelson-io]

Still no resolution for me on this issue. With or without content filters, I always get the message about the model not being safe. Created a new model, new filters, etc.