> ## Documentation Index
> Fetch the complete documentation index at: https://docs.meshai.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Semantic Kernel

> Auto-track LLM calls in Microsoft Semantic Kernel

## Installation

```bash theme={null}
pip install meshai-sdk[semantic-kernel]
```

## Usage

### Option 1: Global Tracking

```python theme={null}
from meshai import MeshAI
from meshai.integrations.semantic_kernel import track_semantic_kernel
from semantic_kernel import Kernel

client = MeshAI(api_key="msh_...", agent_name="my-sk-agent")
client.register(framework="semantic-kernel")

kernel = Kernel()
# ... add AI services to kernel ...

# Enable tracking on the kernel
track_semantic_kernel(client, kernel)

# Use as normal, all LLM calls tracked
result = await kernel.invoke(my_function, input="Hello")
```

### Option 2: Prompt Filter

```python theme={null}
from meshai.integrations.semantic_kernel import MeshAIPromptFilter
from semantic_kernel import Kernel

kernel = Kernel()
# ... add AI services to kernel ...

# Add as a prompt filter
kernel.add_filter("prompt", MeshAIPromptFilter(client))

result = await kernel.invoke(my_function, input="Hello")
# Model and tokens tracked automatically
```

## How It Works

MeshAI integrates via Semantic Kernel's function invocation filter system. The `MeshAIPromptFilter` intercepts prompt render and completion events:

1. Extracts the **model name** from the AI service configuration
2. Extracts **token counts** from the completion result metadata
3. Infers the **provider** from the model or service type
4. Sends the usage event to MeshAI (buffered, non-blocking)

Works with all Semantic Kernel AI connectors: OpenAI, Azure OpenAI, Google, HuggingFace, and others.

## Alternative: Proxy (Zero-Code)

If your Semantic Kernel uses OpenAI or Azure OpenAI, route through the proxy:

```bash theme={null}
export OPENAI_BASE_URL=https://proxy.meshai.dev/v1/openai/k/msh_YOUR_PROXY_KEY
```