dais-polymtl/flock

What it solves

Flock is a DuckDB extension that allows users to perform complex semantic analysis and AI tasks directly within their SQL queries. It bridges the gap between traditional structured data analytics (OLAP) and unstructured data processing using Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG) pipelines.

How it works

Flock integrates LLM capabilities into the DuckDB database engine. It provides a declarative SQL interface where users can call functions for text generation, classification, and embedding generation. It supports multiple AI providers (OpenAI, Azure, Ollama, and Anthropic/Claude) and can handle multimodal inputs including text, images, and audio (via transcription).

Who it’s for

Data analysts and developers who use DuckDB and want to integrate AI-powered semantic analysis, RAG workflows, and multimodal data processing directly into their database workflows without leaving the SQL environment.

Highlights

• Declarative SQL Interface: Use SQL to perform text generation, summarization, and classification.
• Multi-Provider Support: Compatible with OpenAI, Azure, Ollama, and Anthropic/Claude.
• End-to-End RAG: Built-in support for retrieval and augmentation workflows.
• Multimodal Capabilities: First-class support for text, images, and audio transcription.
• LLM Observability: Built-in metrics tracking for token usage, latency, and API call counts.
• WASM Support: Can be run in the browser via DuckDB-WASM for client-side analytics.