Other articles particularly relevant to this article are:

You (as an admin within your Planhat tenant) can choose which AI models are used for "Speed" (quick/simple) and "Advanced" (more complex) tasks within your Planhat tenant (i.e. across your organization).

In many Planhat tenants, these two settings are left as the default "Planhat", which means that Planhat/Pi will automatically apply the most appropriate model, using Planhat's own LLM connections (Planhat's managed models).

However, some enterprises may want to use their own LLM connections, for control and compliance.

You select your choice of model within the "AI Governance Hub" in the Agents Module (Agents Tool). The settings are called "Model selection", under the "Model configuration" heading.

You need to have set up your AI connections in the "Connections" part of the Settings Tool for you to select them as your models. Your own AI provider can be OpenAI, Azure, Anthropic/Claude, Gemini, etc.

Your AI model selection here affects AI functionality across your Planhat OS tenant:

In many cases, leaving these settings on the default "Planhat" is the appropriate choice.

Nevertheless, some enterprise organizations or security teams prefer to use their own LLM connection. The ability to choose your own AI provider gives you control over your AI infrastructure and data governance.

Bringing your own LLM (AI connection) can also reduce the usage of Planhat AI credits, which are consumed when you carry out certain actions within Planhat. However, it's important to note that even if you BYOLLM, using Pi will still consume "Pi Answers" - we discuss this more next.

Pi gives you the ability to chat with AI in Sessions. Your User Type determines the number of "Pi Answers" available. The number of available Pi Answers resets daily. Simple queries are free - they don't count as part of the Pi Answers allowance.

The AI credits and Automation credits that are consumed in Custom Automations are separate from Pi - these credits are not used by Pi.

Even if you BYOLLM, when you carry out complex tasks using Pi, this will consume Pi Answers.

It's important to note that the "cost" of Pi is more than simply AI token processing. On top of connecting to different LLM providers, Pi includes a large amount of complex structure - reasoning etc. - built by Planhat on top of this starting point. Pi is an entire "harness" around the raw LLM - this proprietary "scaffolding" is what makes Pi so powerful. The tool orchestration, process understanding, logic, and so on, is the majority of the value.

Raw token cost is a minority of what is delivered with Pi. Planhat is delivering specialized intelligence, not reselling AI tokens.

There are two stages to this:

Here we'll actually start by explaining stage 2, so you are familiar with the BYOLLM/BYOM settings that are the focus of this article, and then we will go back to stage 1 and show you where you set up connections in Planhat OS.

At the top of the Agents Module (also called the Agents Tool), you will see a part call called "Governance" (formerly called "Analytics").

Mouse over this, and click on the "More" that appears.

This opens up a panel on the right-hand side. (You can click the "x" in the top right of the main Agents Module to close that panel so that the configuration panel has more space.)

In this article, we're focusing on the "Model selection" settings, which you'll find under the "Model configuration" heading, as you can see annotated in the screenshot above.

Clicking on "Model selection" to open up this setting, you'll see there are two dropdown menus where you can select an AI model. The default is "Planhat", but you can change either or both of these to your choice of model.

You can select your own LLM/model for both or either of these.

Here you pick a model optimized for high-volume and relatively simple tasks at a reasonable cost, such as Conversation classification.

Model requirements: at least a 250K-token context window, support for thinking/reasoning mode, and support for structured (strict) JSON responses.

Here you select a model optimized for complex tasks that require more intelligence, such as using Pi to generate HTML artifacts or Custom Automations.

Model requirements: support for tool/function calling, in addition to all requirements listed above for the Speed model. A frontier model is recommended here for the best results.

AI models being selected in these dropdown menus need to be first configured as connections - this is where you connect Planhat to your own external LLM.

To navigate to the connections part of Planhat OS:

[Tags for search purposes: BYO LLM, BYO-LLM]