#About
Instill App introduces an AI Assistant that integrates with Instill Catalog, offering natural language interactions and RAG-based (Retrieval-Augmented Generation) responses for an enhanced user experience.
Instill Applications serve as a product line to demonstrate the seamless integration and synergy between Instill VDP, Instill Model, and Instill Artifact. These applications are tailored to address complex real-world use cases, showcasing the practical applications of these services. The AI Assistant is the first in this series, designed to illustrate the integration and use of Catalogs and Pipelines for intelligent retrieval and interaction with unstructured data.
#AI Assistant Overview
The AI Assistant serves as a user-friendly application that interfaces with Instill Catalog. Before setting up an AI Assistant, users must first create a Catalog in Instill Artifact and upload and process files. Once the Catalog is established, the AI Assistant can interact with the processed files, enabling users to ask questions about the content through natural language queries.
#What is RAG (Retrieval-Augmented Generation)?
RAG is a method that combines retrieval and generation techniques. It uses a retrieval module to extract relevant information from a large dataset and integrates this information into a generation model, enabling more accurate and enriched responses. This approach enhances the ability of the LLM (Large Language Model) to understand queries and generate meaningful answers, positioning RAG as a leading technology in natural language understanding and generation.
In Instill Artifact, users can create custom Catalogs, and within Instill App, they can establish their own AI Assistants. This process provides a user-friendly interface that enables users to interact with the Catalog, quickly verifying and experiencing the benefits of RAG-based systems. Even non-technical users, such as project managers, HR personnel, or sales teams, can easily understand and utilize RAG for basic applications, such as engaging in conversations with documents.
#How It Works
#Catalog Integration
After uploading and processing files in Instill Catalog, users can select an existing Catalog in the AI Assistant to interact with. The selected Catalog serves as the RAG knowledge base, allowing the AI Assistant to seamlessly retrieve relevant information chunks to answer user queries. The AI Assistant also takes historical conversation context into account when responding, ensuring more accurate and contextually aware answers.
#Preset Pipelines
The AI Assistant leverages two preset pipelines to process user queries. These pipelines include predefined prompts for general use (detailed on the Chat page). If the AI Assistant is unable to find sufficient information in the selected Catalog to answer a user's question, it will prompt the user to upload more relevant documents. To enhance the AI Assistant's responses, we recommend uploading documents such as FAQs to the Catalog.
#Limitations
Certain queries, such as greetings, casual conversation, and arithmetic questions, can be answered without the Catalog LLM. However, the AI Assistant may refuse to answer or state that it cannot find relevant information in the Catalog unless related documents (e.g., common Q&A) are uploaded. This is by design, and helps guard against generating inaccurate or misleading information (e.g., hallucinations).
During the Alpha Release period, Instill Cloud Applications have the following limitations:
- Max number of Apps (AI Assistants per namespace):
- Free tier:
10
- Pro tier:
50
- Team & Enterprise tiers:
Unlimited
- Free tier: