How-to · Lesson 2 of 19

Enterprise LLM Configuration and Model Cost Control

Connect a model provider to MagOneAI: name a config, store the API key in KeyVault, load the provider's models, set what each one is allowed to do, test it, and cap cost per token before you create it.

Beginner1:53Learning path · Lesson 2, Set Up

What You Will Learn

Where LLM Configs live in your organization and who can manage them
How to create a new config: name, description, provider, and endpoint URL
How to enter the model's name and add your API key securely
How to enable capabilities like vision and reasoning, and add custom parameters
How to set token limits, image handling, and timeouts for production use
How to choose your organization's default model
How to control which projects can use which LLM
How to track token usage, requests, latency, and errors per config

Step by step guide

Step 1: Open the LLM Configurations Page

From your dashboard, navigate to the LLM Configs section.

The LLM Configs section in the dashboard navigation

Step 2: Start Creating a New Configuration

Click Add Config to create a new LLM configuration.

The Add Config button on the LLM configurations page

Step 3: Understand Supported Model Providers

Mag One AI is a model-agnostic platform. Any model compatible with the OpenAI-style API format is supported, including:

OpenAI
Anthropic
Google Gemini
Self-hosted models

Supported providers, including OpenAI, Anthropic, Google Gemini and self-hosted models

You can connect to any of these providers through a compatible API.

Step 4: Name the Configuration and Add API Key

Provide a clear, descriptive name for your configuration (for example, Fast Model), and enter the corresponding API key for the chosen provider.

API keys are stored securely in KeyVault and are never stored as plain text in the database. Every time you add an API key, it is securely encrypted and managed via KeyVault.

Step 5: Load Available Models from the Provider

Click Load Model to automatically fetch the latest available models from the provider (for example, from ChatGPT via OpenAI).

The Load Model button beside the provider API key field

This will retrieve all current models exposed by the provider.

The list of models returned by the provider

If the provider (e.g., ChatGPT) releases a new model later, it will also appear here when you reload.

Updating to newly released models is as simple as reloading and selecting the desired model.

Step 6: Configure Basic Model Capabilities

Review the available flags and capability options for the selected model.

For example, you can enable Supports Vision if the model can process images.

Model capability flags, including the Supports Vision toggle

You can then refine additional capabilities such as reasoning options.

For a faster configuration, you might set Reasoning to None and Reasoning Effort to None.

Reasoning and Reasoning Effort both set to None

Step 7: Test the Configuration

Before finalizing your LLM configuration, always run a Test Config to ensure everything works correctly.

Test Config results showing the sample request and the model's reply

A sample request is sent (for example, “reply with a single word ‘okay’”), and you can inspect both the request and the response. If the response is as expected (e.g., okay), your configuration is functioning correctly.

Step 8: Adjust Advanced Settings and Create the Configuration

Optionally, configure advanced parameters such as:

Maximum input tokens
Maximum output tokens
Image handling behavior
Token cost per input token
Token cost per output token

Advanced settings: token limits, image handling, and cost per input and output token

After reviewing and adjusting these settings, click Create Config to save your new LLM configuration.