VPT.Einkr.Client 2.0.0

VPT.Einkr.Client

A small, dependency-free .NET client for Einkr's OpenAI-compatible endpoint. It talks to the standard POST /v1/chat/completions and GET /v1/models — the same surface any OpenAI client uses — authenticated with a per-user Einkr API key. Use it when a tool just needs a system + user → text completion without pulling in the rest of Einkr.

• Single request/response — no streaming, no tool calling, no server-side session.
• Depends only on the base class library (System.Net.Http.Json).
• Shaped as a near drop-in for the official OpenAI .NET SDK's ChatClient.CompleteChat.

If you'd rather use the official OpenAI SDK, point it at https://einkr.com/v1 with your Einkr API key as the key — it hits the identical endpoint. This package is the lighter first-party option.

Install

dotnet add package VPT.Einkr.Client

Requires .NET 10.

Quick start

using VPT.Einkr.Client;

var einkr = new EinkrAIClient("Stratum-27B", "vrc_your_api_key");

ChatCompletion response = einkr.CompleteChat([
    new SystemChatMessage("You are a terse assistant."),
    new UserChatMessage("Name three primary colors.")
]);

Console.WriteLine(response.Content[0].Text);
Console.WriteLine($"{response.Usage.CompletionTokens} tokens, finish: {response.FinishReason}");

CompleteChat is synchronous; CompleteChatAsync is the same call with a CancellationToken.

API key

Generate one on Einkr's Profile → API Access tab (it looks like vrc_...). It is sent as a Bearer token. Treat it as a secret — read it from configuration or an environment variable, don't hard-code it.

Choosing a model

The first argument is an Einkr model id. List what a deployment currently offers:

foreach (var m in einkr.GetModels())
    Console.WriteLine($"{m.Id}  (context: {m.ContextWindow})");

An unknown or disabled id fails the first call with a model_not_found error.

Base URL

The constructor's optional third argument is the host root; the client appends /v1 itself:

new EinkrAIClient("Stratum-27B", key);                          // https://einkr.com  (default)
new EinkrAIClient("Stratum-27B", key, "https://einkr.com");     // same
new EinkrAIClient("Stratum-27B", key, "https://localhost:7199");// a local deployment

A URL that already ends in /v1 is accepted without doubling up. The EINKR_API_BASE_URL environment variable overrides the default (but not an explicit argument).

Multi-turn conversations

The client is stateless. To continue a conversation, pass the previous result's History back in — it already contains your messages plus the assistant's reply:

var first = einkr.CompleteChat([
    new SystemChatMessage("You are a helpful assistant."),
    new UserChatMessage("My name is Josh.")
]);

var second = einkr.CompleteChat([
    ..first.History,
    new UserChatMessage("What's my name?")
]);

Thinking / reasoning

enableThinking is bool? and defaults to null — meaning use the model's server-side default. Pass true or false to override it for a single call. Any reasoning the model produces is surfaced on ChatCompletion.Reasoning (never folded into History).

var res = einkr.CompleteChat(
    [new UserChatMessage("Is 91 prime? Think it through.")],
    enableThinking: true);

Console.WriteLine(res.Reasoning);      // the model's working, if any
Console.WriteLine(res.Content[0].Text);// the answer

reasoningEffort ("low" / "medium" / "high" / …) is an optional related knob, sent only when set.

Other options

einkr.CompleteChat(
    messages,
    maxTokens: 256,      // output ceiling for this call
    temperature: 0.2);   // omitted when null -> the model's default applies

Result shape

ChatCompletion carries:

Errors

A non-success response throws EinkrAIException, with the OpenAI error envelope's message where present (otherwise the raw body):

try
{
    var res = einkr.CompleteChat([new UserChatMessage("Hi")]);
}
catch (EinkrAIException ex)
{
    Console.Error.WriteLine(ex.Message); // e.g. "Einkr returned 404: The model '...' does not exist..."
}

Upgrading from 1.x

2.0 retargets from Einkr's own /api/completions to the OpenAI-compatible /v1 surface. Two changes to be aware of:

• ModelInfo is now (string Id, int? ContextWindow) — /v1/models returns string ids, not the old numeric ids.
• enableThinking is now bool? (default null = server default). Existing enableThinking: true / false call sites keep working and now mean "force this for the call."

The EinkrAIClient / CompleteChat / Content[0].Text surface is otherwise unchanged.