CustomProviders

Many hosted APIs are OpenAI-compatible. tryAGI.OpenAI.CustomProviders wraps those endpoints in the same OpenAiClient MEAI adapters, so you can reuse the IChatClient pattern across providers and the IEmbeddingGenerator<string, Embedding<float>> pattern where the provider exposes embeddings.

There are currently 27 factory methods in CustomProviders.

Installation

dotnet add package tryAGI.OpenAI

Provider Table

The factory methods do not hardcode model IDs. The values below are the typical models used in the current docs and tests.

Provider	Factory method	Typical chat model
GitHub Models	`CustomProviders.GitHubModels(token)`	`gpt-4o`
Azure OpenAI	`CustomProviders.Azure(key, endpoint)`	`gpt-4o-mini`
DeepInfra	`CustomProviders.DeepInfra(key)`	`Qwen/Qwen2.5-72B-Instruct`
Groq	`CustomProviders.Groq(key)`	`llama-3.3-70b-versatile`
xAI	`CustomProviders.XAi(key)`	`grok-3-mini`
DeepSeek	`CustomProviders.DeepSeek(key)`	`deepseek-chat`
Fireworks	`CustomProviders.Fireworks(key)`	`accounts/fireworks/models/llama-v3p3-70b-instruct`
OpenRouter	`CustomProviders.OpenRouter(key)`	`meta-llama/llama-3.2-3b-instruct:free`
Together	`CustomProviders.Together(key)`	`meta-llama/Llama-3.3-70B-Instruct-Turbo`
Perplexity	`CustomProviders.Perplexity(key)`	`sonar`
SambaNova	`CustomProviders.SambaNova(key)`	`Meta-Llama-3.3-70B-Instruct`
Mistral	`CustomProviders.Mistral(key)`	`mistral-large-latest`
Codestral	`CustomProviders.Codestral(key)`	`codestral-latest`
Hyperbolic	`CustomProviders.Hyperbolic(key)`	`meta-llama/Llama-3.3-70B-Instruct`
Cohere	`CustomProviders.Cohere(key)`	`command-r-08-2024`
Ollama	`CustomProviders.Ollama()`	`llama3.2`
LM Studio	`CustomProviders.LmStudio()`	user-selected
Cerebras	`CustomProviders.Cerebras(key)`	`llama3.1-70b`
Nebius	`CustomProviders.Nebius(key)`	`Qwen/Qwen2.5-72B-Instruct`
Nvidia NIM	`CustomProviders.Nvidia(key)`	`meta/llama-3.3-70b-instruct`
Ollama Cloud	`CustomProviders.OllamaCloud(key)`	`llama3.2`
MiniMax	`CustomProviders.Minimax(key)`	`MiniMax-M1`
Novita AI	`CustomProviders.NovitaAI(key)`	user-selected
Qwen	`CustomProviders.Qwen(key)`	user-selected
Lepton AI	`CustomProviders.LeptonAI(key)`	user-selected
Cleanlab TLM	`CustomProviders.Cleanlab(key)`	user-selected
SiliconFlow	`CustomProviders.SiliconFlow(key)`	user-selected

Usage

Basic chat

using Microsoft.Extensions.AI;
using tryAGI.OpenAI;

IChatClient chatClient = CustomProviders.DeepSeek(apiKey);

var response = await chatClient.GetResponseAsync(
    "Hello!",
    new ChatOptions { ModelId = "deepseek-chat" });

Embeddings

using Microsoft.Extensions.AI;
using tryAGI.OpenAI;

IEmbeddingGenerator<string, Embedding<float>> generator =
    CustomProviders.DeepInfra(apiKey);

var embeddings = await generator.GenerateAsync(
    ["Hello, world!"],
    new EmbeddingGenerationOptions { ModelId = "BAAI/bge-en-icl" });

Local runtimes

using Microsoft.Extensions.AI;
using tryAGI.OpenAI;

IChatClient ollama = CustomProviders.Ollama();
IChatClient lmStudio = CustomProviders.LmStudio();

CustomProviders vs Dedicated SDKs

Aspect	`CustomProviders`	Dedicated SDK
Installation	`tryAGI.OpenAI`	`tryAGI.{Provider}`
Coverage	OpenAI-compatible chat and embedding paths	Full provider API surface
MEAI support	Reuses OpenAI MEAI adapters	Provider-specific MEAI surface
Provider-specific extras	Limited to OpenAI-compatible endpoints	Provider-native features, auth quirks, and non-chat APIs

Use CustomProviders when the OpenAI-compatible path is enough.

Use a dedicated SDK when you need provider-native capabilities such as:

non-OpenAI endpoints
custom auth semantics
speech, document, or media APIs
provider-specific AIFunction wrappers