Semantic Kernel Integration

Add MeshGuard governance to your Microsoft Semantic Kernel agents with a single line of code. The MeshGuard.SemanticKernel package implements Semantic Kernel's IFunctionInvocationFilter interface, intercepting every function call for policy checks and audit logging — no changes to your existing plugins required.

Installation

bash

dotnet add package MeshGuard
dotnet add package MeshGuard.SemanticKernel

Requires .NET 8+ and Semantic Kernel 1.x.

Quick Start

1. Get Your Credentials

2. Add Governance to Your Kernel

csharp

using Microsoft.SemanticKernel;
using MeshGuard.SemanticKernel;

// Build the kernel with your AI service and plugins
var builder = Kernel.CreateBuilder();

builder.AddAzureOpenAIChatCompletion(
    deploymentName: "gpt-4o",
    endpoint: "https://your-resource.openai.azure.com/",
    apiKey: Environment.GetEnvironmentVariable("AZURE_OPENAI_KEY")!
);

// One-liner: add MeshGuard governance
builder.AddMeshGuardGovernance(options =>
{
    options.GatewayUrl = "https://gateway.meshguard.app";
    options.ApiKey = Environment.GetEnvironmentVariable("MESHGUARD_API_KEY")!;
    options.AgentId = "support-agent-prod";
});

var kernel = builder.Build();

// Import your plugins as usual
kernel.ImportPluginFromType<EmailPlugin>();
kernel.ImportPluginFromType<CalendarPlugin>();

// Every function call is now governed by MeshGuard policy
var result = await kernel.InvokeAsync("EmailPlugin", "SendEmail", new()
{
    ["to"] = "customer@example.com",
    ["subject"] = "Your order update",
    ["body"] = "Your order has shipped!"
});

That's it. Every plugin function call now checks permissions against your MeshGuard gateway before executing, and logs an audit entry after.

How It Works

MeshGuard.SemanticKernel registers an IFunctionInvocationFilter that intercepts the Semantic Kernel function pipeline:

Before invocation — the filter sends a permission check to the MeshGuard gateway, including the function name, plugin name, agent identity, and call arguments. The gateway evaluates the request against your configured policies.
After invocation — regardless of outcome, the filter posts an audit entry recording the function call, the policy decision, execution duration, and (optionally) a summary of the result.
Denied actions — if the gateway denies the call, the filter throws a MeshGuardDeniedException by default. This integrates cleanly with Semantic Kernel's error handling. You can configure it to return a soft denial instead (see ThrowOnDenied below).

Agent → Kernel.InvokeAsync()
         ├── MeshGuard Filter (pre-invocation)
         │     └── POST /v1/check → gateway evaluates policies
         │           ├── ✅ Allowed → continue to function
         │           └── ❌ Denied  → throw MeshGuardDeniedException
         ├── Plugin Function executes
         └── MeshGuard Filter (post-invocation)
               └── POST /v1/audit → log result

Configuration Options

Pass options to AddMeshGuardGovernance() or bind from IConfiguration:

Property	Type	Default	Description
`GatewayUrl`	`string`	—	MeshGuard gateway endpoint (required)
`ApiKey`	`string`	—	API key for gateway authentication (required)
`AgentId`	`string`	—	Unique identifier for this agent instance (required)
`DefaultTrustTier`	`string`	`"standard"`	Trust tier sent with permission checks (`"standard"`, `"elevated"`, `"restricted"`)
`ThrowOnDenied`	`bool`	`true`	When `false`, denied calls return a `MeshGuardDenialResult` instead of throwing
`TimeoutSeconds`	`int`	`5`	HTTP timeout for gateway calls; the function proceeds on timeout if `ThrowOnDenied` is `false`

Binding from appsettings.json

json

{
  "MeshGuard": {
    "GatewayUrl": "https://gateway.meshguard.app",
    "ApiKey": "mg_live_...",
    "AgentId": "support-agent-prod",
    "DefaultTrustTier": "standard",
    "ThrowOnDenied": true,
    "TimeoutSeconds": 5
  }
}

csharp

builder.AddMeshGuardGovernance(
    builder.Configuration.GetSection("MeshGuard")
);

Policy Examples

Define policies in your MeshGuard gateway configuration (YAML). These examples target Semantic Kernel function calls:

Restrict Email Plugin to Trusted Agents

yaml

# Only agents with "elevated" trust can send email
- name: email-trusted-only
  match:
    plugin: EmailPlugin
    function: SendEmail
  rules:
    - condition: agent.trust_tier != "elevated"
      action: deny
      reason: "Email sending requires elevated trust tier"

Rate Limit API Calls

yaml

# Limit any single agent to 100 API plugin calls per minute
- name: api-rate-limit
  match:
    plugin: ApiPlugin
  rules:
    - condition: rate_limit("per_agent", 100, "1m")
      action: deny
      reason: "API call rate limit exceeded (100/min)"

Block File Access to Sensitive Directories

yaml

# Prevent file operations targeting sensitive paths
- name: block-sensitive-dirs
  match:
    plugin: FilePlugin
  rules:
    - condition: >
        args.path starts_with "/etc" or
        args.path starts_with "/var/secrets" or
        args.path contains ".ssh"
      action: deny
      reason: "File access to sensitive directories is prohibited"

Advanced Usage

Custom Context Enrichment

Add business context to every governance check by implementing IMeshGuardContextEnricher:

csharp

public class TenantContextEnricher : IMeshGuardContextEnricher
{
    private readonly ITenantService _tenants;

    public TenantContextEnricher(ITenantService tenants)
        => _tenants = tenants;

    public async Task EnrichAsync(MeshGuardContext context)
    {
        var tenant = await _tenants.GetCurrentAsync();
        context.Properties["tenant_id"] = tenant.Id;
        context.Properties["tenant_plan"] = tenant.Plan;
        context.Properties["environment"] = "production";
    }
}

// Register it
builder.Services.AddSingleton<IMeshGuardContextEnricher, TenantContextEnricher>();

Enriched properties are available in policy conditions as context.tenant_id, context.tenant_plan, etc.

Per-Plugin Governance Policies

Override governance options for specific plugins when you need different trust tiers or timeout behavior:

csharp

builder.AddMeshGuardGovernance(options =>
{
    options.GatewayUrl = "https://gateway.meshguard.app";
    options.ApiKey = Environment.GetEnvironmentVariable("MESHGUARD_API_KEY")!;
    options.AgentId = "multi-plugin-agent";

    // Elevated trust for the calendar plugin
    options.PluginOverrides["CalendarPlugin"] = new PluginGovernanceOptions
    {
        TrustTier = "elevated",
        TimeoutSeconds = 10,
    };

    // Restricted trust for file operations
    options.PluginOverrides["FilePlugin"] = new PluginGovernanceOptions
    {
        TrustTier = "restricted",
        ThrowOnDenied = true,
    };
});

Integration with Entra ID for Agent Identity

Bind agent identity to Microsoft Entra ID (Azure AD) managed identities for zero-secret deployments:

csharp

builder.AddMeshGuardGovernance(options =>
{
    options.GatewayUrl = "https://gateway.meshguard.app";
    options.AgentId = "support-agent-prod";

    // Use Entra ID managed identity instead of a static API key
    options.UseEntraIdAuthentication(
        tenantId: "your-tenant-id",
        clientId: "your-app-client-id"
    );
});

The SDK acquires tokens via Azure.Identity.DefaultAzureCredential and sends them as bearer tokens to the gateway.

Feeding Audit Logs to Microsoft Purview

Route MeshGuard audit events to Microsoft Purview for unified compliance reporting:

csharp

builder.AddMeshGuardGovernance(options =>
{
    options.GatewayUrl = "https://gateway.meshguard.app";
    options.ApiKey = Environment.GetEnvironmentVariable("MESHGUARD_API_KEY")!;
    options.AgentId = "compliance-agent";

    // Forward audit logs to Purview
    options.AuditSinks.Add(new PurviewAuditSink(
        purviewAccountName: "your-purview-account",
        collectionName: "ai-agent-governance"
    ));
});

Audit events appear in Purview's Data Map with full lineage — linking agent identity, function call, policy decision, and outcome.

MeshGuard .NET SDK on GitHub — Source code, issues, and contribution guide
Governing Microsoft Copilot Agents — Patterns for governing Copilot-based agents with MeshGuard
MeshGuard vs Microsoft Purview — Comparison of runtime governance vs compliance monitoring

Next Steps

Policy Configuration — Define what your agents can do
Audit Logging — Query and export audit trails
Generic HTTP Integration — For custom .NET or non-SK workloads
API Reference — Direct gateway API access

Semantic Kernel Integration ​

Installation ​

Quick Start ​

1. Get Your Credentials ​

2. Add Governance to Your Kernel ​

How It Works ​

Configuration Options ​

Binding from appsettings.json ​

Policy Examples ​

Restrict Email Plugin to Trusted Agents ​

Rate Limit API Calls ​

Block File Access to Sensitive Directories ​

Advanced Usage ​

Custom Context Enrichment ​

Per-Plugin Governance Policies ​

Integration with Entra ID for Agent Identity ​

Feeding Audit Logs to Microsoft Purview ​

Related ​

Next Steps ​

Semantic Kernel Integration

Installation

Quick Start

1. Get Your Credentials

2. Add Governance to Your Kernel

How It Works

Configuration Options

Binding from appsettings.json

Policy Examples

Restrict Email Plugin to Trusted Agents

Rate Limit API Calls

Block File Access to Sensitive Directories

Advanced Usage

Custom Context Enrichment

Per-Plugin Governance Policies

Integration with Entra ID for Agent Identity

Feeding Audit Logs to Microsoft Purview

Related

Next Steps