Create an AI Assistant

Preamble's AI Assistants allow organizations to easily configure guardrails for AI applications and share them with their team.

Initial AI Assistant Window

Follow these few steps to quickly deploy your own AI Assistant

Step 1

Create a name

Create a name for your AI Assistant. We recommend making the names relevant to the use cases, so your team will know how to best use the particular AI Assistant.

Choose a Guardrail Model that will enforce your policies (evaluation model)

• OpenAI Models

  • gpt-4o

  • gpt-40-mini

  • o1-min

  • o3-mini

• Mistral AI (These models have roughly a 24K word context window size.)

  • mistral-large-latest

  • mistral-small-latest

  • open-mistral-7b

  • open-mixtral-8x7b

• Anthropic

  • Claude 3.5 Sonnet

  • Claude 3.5 Opus

  • Claude 3.5 Haiku

• Google

  • Gemini 1.5 Flash

  • Gemini 1.5 Flash 8B

  • Gemini 1.5 Pro

  • Gemini 2.0 Flash Exp

• Custom Models

  • Currently limited to models with similar API formats as OpenAI & Anthropic

When this step is complete, click on the 'Save and Continue' button in the top left hand corner.

*Request new models by contacting us - [email protected]

Select an AI (LLM)

AI systems available to use as your underlying AI Assistant:

• OpenAI Models

  • gpt-4o

  • gpt-40-mini

  • o1-min

  • o3-mini

• Mistral AI (These models have roughly a 24K word context window size.)

  • mistral-large-latest

  • mistral-small-latest

  • open-mistral-7b

  • open-mixtral-8x7b

• Anthropic

  • Claude 3.5 Sonnet

  • Claude 3.5 Opus

  • Claude 3.5 Haiku

• Google

  • Gemini 1.5 Flash

  • Gemini 1.5 Flash 8B

  • Gemini 1.5 Pro

  • Gemini 2.0 Flash Exp

• DeepSeek (USA based)

  • R1

  • V3

• Custom Models

  • Currently limited to models with similar API formats as OpenAI & Anthropic

Step 2

Create default violation responses

This message will be shown in the AI Assistant page when a policy violation has occurred.

• We recommend directing your users to review the policies you have enabled if there is a violation and your contact information

  • i.e. "Your input was flagged as violation to our company policy. If you feel that this was a mistake, contact your administrator."

Create a default initial chat message for your AI Assistant

This will be the initial message your users will see in the AI Assistant chat window. We recommend adding a disclaimer and leading questions to help your users get started.

• i.e. "You are using a controlled and monitored AI system. This assistant can help you with writing project proposals."

When this step is complete, click on the 'Save and Continue' button in the top left hand corner.

Step 3

Select your policies to add to your guardrails

You need to select at least (1) Input and (1) Output policy for your AI Assistant. You can use the same policy for both the input and output if you prefer.

Input Policies

This controls the input text/prompts that users are allowed to submit.

Output Policies

This controls the output text from the AI model you selected in the previous window.

Policy Selection options

• My Policies

  • These are the policies that are private to your organization.

• Public Policies

  • These are policies that have been shared by Preamble and other organizations.

  • These are the initial policies you will see when configuring your AI Assistant.

When this step is complete, click on the 'Save Configuration' button in the top left hand corner.

Step 4

After you save your configuration, you will now be directed to the AI Assistant page to use your new configured AI Assistant.

From the drop down menu in the top left hand corner, select the AI Assistant you would like to use.

Once you click on the Assistant to use, you will see the initial message that you configured in the window.

Begin using the AI by entering a question or request.

*The AI Assistant only allows free trial users to control Input OR Output.