Admin Settings

Manage runtime application configuration through the Admin Settings UI without restarting the application.

Overview

Admin Settings provides:

• Runtime Configuration - Change settings without redeploying or restarting
• LLM Provider Management - Configure and hot-swap AI providers at runtime
• Change History - Full audit trail of all setting changes with user attribution
• Encrypted Storage - Sensitive values (API keys, endpoints) encrypted in the database
• Role-Restricted - Only Admin users can access settings

Accessing Admin Settings

1. Log in as an Admin user
2. Navigate to Admin Settings in the navigation menu (/admin-settings)
3. Configure settings across the available tabs

Note

Admin Settings is only visible to users with the Admin role or Super Admin flag.

Configuration Tabs

General Settings

Setting	Description	Example
Base URL	Application URL for notification links	https://yourdomain.com

The Base URL is used to generate clickable links in Teams/Slack notifications that take users to the Beacon UI (the React app is served at the root URL).

AI Configuration

Configure the LLM provider for AI-powered features (documentation generation, natural language alerts).

Setting	Description	Required
Provider	LLM provider (OpenAI, Anthropic, AzureOpenAI, Bedrock)	Yes
API Key	Authentication key for the provider	Yes
Endpoint	Custom API endpoint URL	No
Region	AWS region (Bedrock only)	Bedrock only
Model	Primary model name	Yes
Fast Model	Lightweight model for quick operations	No
Max Concurrent Requests	Parallel request limit	No (default: 50)
Tokens Per Minute	Rate limit	No (default: 80,000)
Requests Per Minute	Rate limit	No (default: 1,000)
Monthly Budget	Cost cap in USD	No (default: $100)

Caution

API keys are encrypted before storage and masked in the UI. They are never exposed in plain text after saving.

Change History

View a complete audit log of all settings changes:

• Setting Key - Which setting was changed
• Old Value / New Value - Previous and new values (masked for sensitive fields)
• Changed By - User who made the change
• Changed At - Timestamp

Hot-Swap LLM Providers

Admin Settings can change the LLM provider at runtime without restarting the application.

How It Works

1. Admin updates LLM settings via the UI
2. AppSettingsService saves encrypted values to the database
3. LlmProviderManager receives the update via ILlmConfigurationUpdater
4. A new provider instance is created with the updated configuration
5. DelegatingLlmProvider (the proxy injected throughout the app) automatically delegates to the new provider
6. All subsequent AI requests use the new provider immediately

Architecture

AppSettingsService.SaveSettingsAsync()
    │
    ├── Save to database (encrypted)
    ├── Invalidate cache
    ├── Update BeaconConfiguration singleton
    │
    └── ILlmConfigurationUpdater.UpdateConfiguration()
            │
            └── LlmProviderManager
                    ├── Mutate LlmConfiguration singleton
                    └── Recreate ILlmProvider via factory
                            │
                            └── DelegatingLlmProvider (proxy)
                                    └── All consumers use new provider

Example: Switching from OpenAI to Anthropic

1. Go to Admin Settings > AI Configuration
2. Change Provider to Anthropic
3. Update API Key to your Anthropic key
4. Change Model to claude-3-5-sonnet-20241022
5. Click Save
6. AI features immediately use Anthropic — no restart needed

Initial Configuration

Settings can be pre-configured in two ways.

Via appsettings.json (Startup Defaults)

{
  "Beacon": {
    "BaseUrl": "https://yourdomain.com",
    "LLM": {
      "Provider": "OpenAI",
      "ApiKey": "sk-your-api-key",
      "Model": "gpt-4o"
    }
  }
}

On startup, if Admin Settings in the database are empty, values from appsettings.json are used as defaults. Once saved via the Admin Settings UI, database values take precedence.

Via First-Run Setup

After creating the super admin during first-run setup, navigate to Admin Settings to configure the LLM provider and other settings.

Database Storage

Settings are stored in two tables.

app_settings

Column	Type	Description
key	string	Setting identifier (e.g., LLM.ApiKey)
value	string?	Setting value (encrypted if sensitive)
category	string	Grouping (General, LLM)
is_sensitive	bool	If true, value is encrypted

app_setting_history

Column	Type	Description
setting_key	string	Which setting changed
old_value	string?	Previous value (*** if sensitive)
new_value	string?	New value (*** if sensitive)
changed_at	DateTime	When the change occurred
changed_by_user_id	string?	Who made the change

Programmatic Access

Reading Settings

public class MyService
{
    private readonly IAppSettingsService _settingsService;

    public MyService(IAppSettingsService settingsService)
    {
        _settingsService = settingsService;
    }

    public async Task DoSomethingAsync()
    {
        var settings = await _settingsService.GetSettingsAsync();

        var baseUrl = settings.BaseUrl;
        var llmProvider = settings.LlmProvider;
        var llmModel = settings.LlmModel;
    }
}

Saving Settings

var settings = await _settingsService.GetSettingsAsync();
settings.BaseUrl = "https://new-domain.com";

await _settingsService.SaveSettingsAsync(settings, userId: currentUser.Id);

Viewing History

var history = await _settingsService.GetHistoryAsync();

// Filter by setting key
var llmHistory = await _settingsService.GetHistoryAsync(key: "LLM.ApiKey");

Implementing ILlmConfigurationUpdater

If you build custom services that need to react to LLM configuration changes, implement ILlmConfigurationUpdater:

public interface ILlmConfigurationUpdater
{
    void UpdateConfiguration(AppSettingsData settings);
}

This interface lives in Beacon.Core so that AppSettingsService can call it without depending on Beacon.AI. The implementation (LlmProviderManager) lives in the AI project.

Caching

Settings are cached in memory for 1 hour to minimize database queries. The cache is invalidated immediately when settings are saved via the Admin Settings UI.

MCP Settings

MCP server behavior has its own admin page at /mcp-settings, organised into four tabs — Pre-prompt (SQL-generation system prompt, global instruction), Tool descriptions (overrides applied to the live tools/list response), Guardrails (max row limit, read-only enforcement, PII detection with custom patterns, and the learning knobs), and Context preview (see the grounding context for a project exactly as the tools assemble it).

See the MCP Server Guide for details. Additional retrieval, exemplar, and eval settings that aren’t on this page are settable through PUT /beacon/api/mcp/settings — see the Knowledge Base guide. Like LLM settings, MCP settings apply immediately — no restart required.

See Also

• AI Integration - AI features that use LLM configuration
• MCP Server - MCP settings, guardrails, and the learning loop
• Knowledge Base & Grounding - Retrieval, exemplar, and eval settings
• User Management - Role-based access to Admin Settings
• Configuration Guide - Startup configuration reference