MOTIFIC POLICIES

Comprehensive, customizable set of policy controls

Motific provides the following policy controls for each user and model interaction:

• Adversarial and harmful content policy
• Malicious URL protection policy
• Off-topic detection
• Code detection
• Personal data shield
• Content ethics and safety
• Token budgets
• Hallucination detection

Visit our documentation for more information.

Knowledge bases

Simplified, automated configurations with RAG

Simplified Retrieval-Augmented Generation (RAG) configuration and usage
Motific enables you to create knowledge base (KB) configurations so that the model’s response gets relevant contextual information from the data sources. Some key features include:

• Fully-managed ingestion: Motific abstracts away the complexity of connecting with popular data sources such as Microsoft Sharepoint and Websites. Motific can ingest data on a customer defined frequency and store it for accurate retrievals in a vector database.
• Fully-managed retrieval: When a knowledge base is attached to an assistant (or Motif), Motific automatically retrieves the most relevant content for the user query, and sends it to the model for use in answering the query.
• User Access Control: For data sources with granular user access controls, Motific ensures that a user’s query is answered only using data that the user has access to.

Prompt Intelligence

Get data-driven prompt insights

Motific provides insights to organizations on how GenAI users benefit from the AI assistants by looking at the actual work done via the users prompts/ inputs. This data can help with optimizing your AI assistant usage. Motific provides the following usage insights:

• Monitor, detect and control usage of unauthorized LLMs and GenAI services.
• Tasks categories requested by each prompt/user input, such as content creation, question & answering, data analysis, etc.
• The tasks that are most often requested by the user, the token usage trends for each requested task
• The time saved by using an AI assistant for a particular task, and comparisons with different tasks.
• What-if model analysis that allows you to verify if the best model was selected for a given Motif according to the type of tasks users leverage. After comparison with other models, you can choose to reconfigure the Motif to use another model to reduce cost or enhance performance or accuracy.