Post-installation problem with model settings

[Zzegfried]

I'm using Visual Studio 2026 with latest Ollama and Gemma4:e4b model on Windows 11 Pro. After installing LocalPilot extension, the chat repeatedly failed with:

"[LocalPilot Error] Could not reach Ollama: Response status code does not indicate success: 404 (Not Found)."

After a lot of digging I found that the logs indicated that it was trying to connect to the llama model even though I don't have it installed and even though the LocalPilot settings dialog showed Gemma4 configured for all interactions. The chat function only worked after I explicitly saved the settings. It seems to me that when installing the extension it should be initially configured to use the Ollama default model not always llama.

BTW, thank you for sharing this tool. 👍

[JalpeshVadgama]

@Zzegfried Thanks for the update. What I will do it if you have not done settings for the ollama models. It will give you a message that settings are required.

[JalpeshVadgama]

@Zzegfried - With new release 1.7 I have fixed that issue in the following commits - 7446769

So now it will automatically fetch the models available instead of hardcoded values on the first run. And if still models not found then it will give nice message on chat window to configure models under settings. So we can close this issue now

[Zzegfried]

@JalpeshVadgama Thank you.

BTW I gave up using it after many crashes that occurred immediately after changing models in the settings. I also got a lot of protocol errors while attempting to chat with most models I tried. Have you tested it with VS 2026?

[JalpeshVadgama]

@Zzegfried Yes, I am doing testing on Visual Studio 2026 community edition only. Please test it and let me know if you still got the problem. Please enable the log and send log to me . Also, I would recommend to use code centric models like qwen-coder, code llama etc.

If you provide me list of models that you have tried and your PC configuration that would help me get this extension

[Zzegfried]

@JalpeshVadgama I retried with LP version 1.7.0.0 and got the same error when attempting to chat using codellama:latest ...

[LocalPilot Error] Could not reach Ollama: Response status code does not indicate success: 400 (Bad Request).

Log file attached:
localpilot.log

Note that LP successfully connects to Ollama. It just fails to process any prompts.
I am running Visual Studio 2026 Enterprise v18.4.3

[JalpeshVadgama]

HI @Zzegfried

While looking at Log you provided I have intially found the following things

Root Cause Analysis
The sequence of events in your logs reveals two primary issues:

Model Mismatch (404 Not Found):

At 07:56:56, the system attempts to run llama3:8b.

At 08:25:22, it switches to Qwen3-Coder-Next:latest.

At 08:46:57, it attempts gemma4:e4b.

Conclusion: The 404 error occurs because Ollama does not recognize these specific model names or tags. While you may have a "Qwen" or "Llama" model, the exact tag (e.g., :8b or :latest) must match what is currently downloaded in your local Ollama library.

Context & Sync Timeouts (Task Canceled):

The logs show multiple TaskCanceledException errors during "Embedding" and "RAG" indexing.

Conclusion: Your machine is likely struggling to handle the large context being sent (up to 29,072 tokens) while simultaneously indexing 242 files. This causes the internal HTTP client to time out before the local server can respond

So here are potential fixes

1. Verify Model Tags
Open a terminal and run ollama list. Ensure the model name used in your settings matches the "NAME" column exactly.

If you see llama3, but your log shows llama3:8b, you must either rename the model or update your configuration to remove the :8b tag.

Note: "Qwen3" and "Gemma4" are likely typos in the configuration; you likely mean qwen2.5-coder or gemma2

2. Reduce Resource Contention
The logs show "Starting parallel differential sync" immediately before the network errors.

Wait for indexing to finish: Let the RAG indexer complete its work before sending a chat prompt.

Lower Context Allocation: Your system is trying to allocate ~29k tokens. If you are on a system with less than 16GB of VRAM, try lowering the context limit in your settings to 8192 to prevent timeouts.

If you could provide me a your laptop or pc configuration that would be great as I can see ollama is struggling with providing response. If you could try the smaller prompts and see its working of not. I would highly recommend to use smaller model if your pc does not support higher models

I am digging more into logs and trying to reproduce the same on my machine I will get back to you on this once I found something but this is my primary investigation

[Zzegfried]

Hi @JalpeshVadgama

I don't seen anything wrong with the model names and at no time or place did I alter the names. I ran "ollama ls" and validated all models with "ollama pull". In LP settings I have only selected the models from the drop-down list. I have indeed pulled Gemma4:e4b and Qwen3-Coder-Next:latest. The only prompt I have made so far is "Hello" and it fails on that.

I am using a machine with an Intel i7-13700H using integrated GPU only. The machine has 64GB memory. At this point I am just experimenting to see if it's even possible to run any model productively on this hardware. I was using Copilot free tier but we all know what has happened with that recently! 😋

[JalpeshVadgama]

@Zzegfried I have upgraded version to 1.8 and done lots of changes related to performance both indexing and UI Level on this PR

#34
#33

So hope fully it should be good now. Also Now if the models for embedded text for indexing is not there so upgrade it to 1.8 and see if you get problem or it got resolved by the changes

[Zzegfried]

@JalpeshVadgama I am getting the same error. have uninstalled 1.7, deleted the LP settings.json file, and reinstalled. I set the model for all options to "codellame:latest", tested the connection (it's OK) and typed "Hello" in the chat. I get:

[LocalPilot Error] Could not reach Ollama: Response status code does not indicate success: 400 (Bad Request).

Logs:

[06:36:28.285] [INFO] [AGENT] [RAG] Solution changed. Clearing index for D:\GitHub\K______\ManufacturingV2
[06:36:28.291] [INFO] [AGENT] [RAG] Starting parallel differential sync...
[06:36:28.370] [INFO] [AGENT] [RAG] Indexing 252 new or modified files...
[06:36:34.173] [INFO] [AGENT] [Agent] Registered 12 native tools for Ollama (ReadOnly: False)
[06:36:56.651] [INFO] [AGENT] [Agent] Turn 1: Running model codellama:latest with context=36157
[06:36:56.691] [ERROR] [OLLAMA] [Ollama] Core Network Error
[Exception] HttpRequestException: Response status code does not indicate success: 400 (Bad Request).
[StackTrace]    at System.Net.Http.HttpResponseMessage.EnsureSuccessStatusCode()
   at LocalPilot.Services.OllamaService.<StreamChatWithToolsAsync>d__15.MoveNext()

I guess this is something specific to my machine since I was not getting this when I first tried LP, but once it happened there appears no way to avoid it. I forget now where i got additional logs but the HTTP 400 is related to a protocol error related to they payload being sent to Ollama. I'm using Ollama 0.22.1 (also happened on 0.21.2)

[JalpeshVadgama]

@Zzegfried I found that issue is with indexing so what I have done is instead of having file based indexes I have moved to sql light db that will be out of memory and that will be quite a performance boost.

Also from the logs whatever you provided I have found that may be you are using old version of ollama please do upgrade that and also use models that supports agentic workflow like qwen-coder

I want to make this extension awesome and thank you all the feedback till now. Please keep supporting

[Zzegfried]

@JalpeshVadgama The good news is I'm seeing no more errors. The not so good news is that I haven't found a local model that can run on my system without timing out in 5 minutes. I tried Qwen2.5-Coder:7B with my "Hello!" prompt and it ran for 5 minutes and timed out. I assume at this point the problems are all mine. Thank you for chasing this up. If you need me to try anything else, let me know.