Dust

Build powerful workflows on top of LLMs and Semantic Search. Deploy to API or use directly from DUST. Access versions, cache model interactions, and connect to Data Sources.

Features

• Chained LLM Apps: Chain arbitrarily between calls to models, code execution and queries to external services.
• Multiple Inputs: Iterate on LLM app design on several inputs simultaneously to avoid overfitting.
• Model Choice: Design against models served by OpenAI, Cohere, AI21 and more. Switch models seamlessly.
• Version History: Access iterations, model outputs and few-shot examples saved automatically.
• Caching: Speed up iterations and reduce costs with cached model interactions.
• Easy Deployment & Use: Deploy to an API endpoint or use directly from DUST.
• Data Sources: Fully managed semantic search engines you can query from your workflows.
• Community Example Apps: Discover apps created by the community to get started with DUST.

Use Cases:

• Semantic Search for IPCC AR6 Report: Use semantic search to answer questions about the IPCC AR6 report.
• Wedding Thank You Notes: Solve the blank page problem for wedding thank you notes.
• Web Search Assistant: Answer questions with high factual accuracy by searching online and compiling responses based on content from websites.
• Generating Code for Maths Questions: Generate code to answer maths questions.
• Teaching LLM to Teach Itself: Teach LLM to teach itself new tasks by prompting itself with few-shot examples from high-level task descriptions.

DUST is a powerful tool for anyone looking to build and deploy language model apps. It offers a wide range of features, easy deployment, and access to managed data sources.