Help/Bug: Error when trying to run completion

[RiaGruen]

Setup:

• Visual Studio 2026
• FreeAIr 4.12.2
• LM Studio 0.4.20.1, running on localhost

Project language:

• VB.Net

I did manage to make embeddings, however, only two are created.

When I open the chat window and try to run anything against the LLM, I get the following error:

Service request failed.
Status: 400 (Bad Request)
bei OpenAI.ClientPipelineExtensions.ProcessMessage(ClientPipeline pipeline, PipelineMessage message, RequestOptions options) in C:\projects\github\openai-dotnet\src\Custom\Internal\ClientPipelineExtensions.cs:Zeile 51.
bei OpenAI.Chat.ChatClient.CompleteChat(BinaryContent content, RequestOptions options) in C:\projects\github\openai-dotnet\src\Generated\ChatClient.cs:Zeile 28.
bei OpenAI.Chat.ChatClient.<>c__DisplayClass12_0.b__0() in C:\projects\github\openai-dotnet\src\Custom\Chat\ChatClient.cs:Zeile 228.
bei OpenAI.SseUpdateCollection1.<GetRawPages>d__10.MoveNext() in C:\projects\github\openai-dotnet\src\Utility\SseUpdateCollection.cs:Zeile 71. bei System.ClientModel.CollectionResult1.d__1.MoveNext()
bei FreeAIr.BLogic.Reader.LLMReader.d__10.MoveNext() in C:\projects\github\FreeAIr\FreeAIr\BLogic\Reader\LLMReader.cs:Zeile 147.

As this directory doesn't even exist on my machine nor any of the mentioned files, I can't even try to fix it on my own. Sorry for that.

LM Studio doesn't log anything, so it's failing to get to the endpoint in the first place.

Any ideas or pointers?

[lsoft]

Could you check a KoboldCpp instead of LM Studio? Also, it is very helpful to know what model you are using.

So, help me to repro this and I try to fix this.

[RiaGruen]

No, I can't use koboldcpp, it has to be LM Studio.

The request doesn't even get to the endpoint, so the model is irrelevant to the bug. The endpoint ist localhost:1234/v1, which is OpenAI-compatible.

[RiaGruen]

As LM Studio returns a 400 immediately, I presume you've never tested it on other LLM providers? I guess, you've not implemented the engine calling correctly.

In that case, FreeAIr is unusable to anybody except for you. Definitely not for me.

[lsoft]

I'll take a look later, I'm busy with RAG now.

[lsoft]

Looks like I found the issue. I got 400 if I set model name wrong.

Try to copy identifier

and then set it here

then I started new chat and the chat window is working fine:

but when a model name is invalid:

Do not forget to start a NEW chat after settings the model name.

Waiting for reply.

[RiaGruen]

Check the error message again, please, yours is different from mine. 400s can come from various problems.

[lsoft]

I tried to build embeddings via lms, and I succeeded.

As I'm not able to reproduce it, I can't provide any help or tips, sorry for that.

[RiaGruen]

This bug is not about embeddings, but about getting a 400 (Bad Request) when trying to run interference.

Quite obviously my error message is a Status message: Service Request failed.

Yours in the screenshot above is a HTTP one: No models loaded.

[lsoft]

@RiaGruen the problem was in mcp tools json serialization. I can't attach binaries here, but you can build from #65 and verify. or just wait few days for publishing in VS marketplace.

thank for reply.

[lsoft]

hopefully fixed in 4.3.0. feel free to reopen if it's not.