Closed as not planned
Closed as not planned
Feature
Description
Hello team,
We're currently using Microsoft.SemanticKernel.Connectors.Ollama
(v1.54.0-alpha) to integrate local LLMs via Ollama. While the experience is great overall, we noticed that the current extension methods:
AddOllamaEmbeddingGenerator(...)
AddOllamaChatCompletion(...)
do not allow passing a custom HttpClient
, which makes it impossible to configure critical options such as Timeout
, custom headers, retry policies, or advanced diagnostics.
This is a limitation especially for local LLMs like llama3.2:3b
or mistral
, which may take longer than 100 seconds on first generation, leading to TaskCanceledException
.
We kindly suggest either:
- Adding overloads that support injecting a preconfigured
HttpClient
- Or allowing DI via options/configuration pattern
This change would align with .NET best practices and enable full control over networking behavior when using Ollama.
Thank you