opencode-plugboard

Appearance

Semantic Kernel ​

Microsoft's Semantic Kernel, through its OpenAI connector.

Connect (Python) ​

Give the connector your own AsyncOpenAI client pointed at plugboard, and don't also pass api_key or base_url:

py
def make_service() -> OpenAIChatCompletion:
    # The client is passed in; do not also pass api_key or base_url to the service (they conflict).
    client = AsyncOpenAI(base_url=BASE, api_key=KEY)
    return OpenAIChatCompletion(ai_model_id="fast", async_client=client)

A script with a text reply, a stream and a plugin function call is in examples/semantic-kernel/:

sh
cd examples/semantic-kernel
PLUGBOARD_KEY=<app key> uv run chat.py

Connect (.NET) ​

csharp
builder.Services.AddOpenAIChatCompletion(
    modelId: "fast",
    endpoint: new Uri("http://127.0.0.1:4010/v1"),
    apiKey: Environment.GetEnvironmentVariable("PLUGBOARD_KEY")!);

What works ​

Chat Completions with streaming, and kernel plugin functions with FunctionChoiceBehavior.Auto (automatic or manual invocation).

Limits ​

  • Semantic Kernel sends tool names prefixed with the plugin name (weather-get_weather); expect that name in tool calls.
  • Set max_completion_tokens high enough (the example uses 1024) when the model reasons.